Porpoise Optimized Deep Learning Based Multi-Hop Routing Protocol for Efficient Data Integration in WSN

Porpoise Optimized Deep Learning Based Multi-Hop Routing Protocol for Efficient Data Integration in WSN | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Porpoise Optimized Deep Learning Based Multi-Hop Routing Protocol for Efficient Data Integration in WSN Ramesh. D, T. Jaya This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7799400/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Rise of Wireless Sensor Networks (WSNs) for applications ranging from environmental monitoring to healthcare, defence, and disaster management is an essential technology in the recent trends. Resource constraints such as energy resources, redundant data transmission, routing inefficiencies, and security vulnerabilities prime to deprived performance. To overcome these restrictions, a Bioinspired algorithm based on Porpoise Optimization Algorithm (POA) optimized deep learning-based multi-hop routing protocol designed for efficient data integration in WSNs. The proposed framework integrates the feature extraction capability of Stacked Auto encoders (SAEs) with the global and local search efficiency of POA to optimize cluster head selection, multipath routing, and data aggregation. In this approach, nodes are clustered into clusters based on remaining energy, node density, and distance, with cluster heads selected using a POA enhanced latent space mechanism. The Porpoise Optimization Algorithm - Stacked Auto encoders (POA-SAE) model compressed the high dimensional sensor data into compact latent representations, thereby minimizing communication overhead, redundancy of data and improving data accuracy. The integration of POA and SAE ensures balanced energy consumption across nodes by dynamically identifying the optimal routing paths through encircling and spiral search strategies inspired by porpoise foraging behaviour. Simulations are carried out in NS3 Whereas performance evaluation is carried out in MATLAB environments to demonstrate that the proposed POA-SAE protocol significantly outperforms conventional approaches such as LEACH, PSO, and GA. Performance metrics including residual energy, clustering time, packet delivery ratio, latency, and data aggregation efficiency expose substantial improvements in scalability, stability, and network lifetime. Specifically, POA-SAE achieves faster convergence, higher residual energy retention, and superior cluster distribution, while effectively reducing transmission delays and redundant packets. Scalability Energy Efficient Reliability Network life time Data aggregation Redundant data Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction Wireless Sensor networks comprised of multiple interconnected nodes, deployed in areas that are inaccessible to humans without the need for any physical media. Sensor nodes assist other nodes understand and interpret information. Lifetime of the network is indirectly proportional to its energy usage, results in increased lifespan because of its lower energy consumption [ 1 ]. A system that is only productive for a brief amount of time poses a severe hazard. Techniques for overloading short-lived networks include nap cycle arrangement, information reduction, and stereo recovery [ 2 ]. All objects in the network can be logically or physically communicated with by community nodes; the application focus on the subsequent social actor actions. The rational function of the links primarily determines the logical architecture. Sometimes it's random, and other times it's a plan. Centralized approaches are suitable as a system's computational energy capabilities rely mostly on a single magnetic generator [ 3 ]. This specific unit is in charge of managing, coordinating, and processing the sensed information pursuits in such cases. Sensor nodes deployed in a hostile environment, security concerns, energy usage, data secrecy, delay in packet transmission between two sensor nodes, trust worthiness of data, and data aggregation become crucial [ 4 ]. Sensor system can achieve emergent behaviour in which nodes communicate independently and coordinate on their own. The objective is to accomplish things that go beyond what each node alone can do. WSN security is used in a diverse situation, including medical, military, civilian duties, and disaster relief. Furthermore, in order to collect data sensor nodes transmission of packets from sensed sensor node to respective Base Station (BS) is essential in WSN. Another important aspect of WSN that could be used to enhance the performance of computing is parallelism [ 5 ]. A good routing procedure is necessary for the efficient transfer of information, and from the perspective of the consumer, fake or even delayed information is insufficient. Routing protocols are divided into a few groups based on community structure and process technique. Routing protocols can be categorized as location-dependent, Layered, or possibly based on the design of the system. Despite their emphasis on security, all routing processes have pros and cons. To manage the integrity, confidentiality, and authenticity of sensed data, a protective framework is required [ 6 ]. Both efficacy and communal life expectancy were linked to protection; Recently, a lot of research has been done to develop dependable and efficient routing methods. Spoofing or even altering the information path, Sybil encounter, choosy forwarding, wormhole encounter, sinkhole encounter, and HELLO foods encounter are some of the many attacks that occur throughout the system-level process [ 7 ]. In order to achieve energy effectiveness, clustering systems may also be preferred only in multihop information exchange. To achieve greater energy efficiency and network scalability, sensor nodes are arranged directly into clusters. The BS's cluster heads (CHs) act as routers, sending data from CHs located farther away. This is merely due to the fact that certain nodes, which have a wealth of web resources, are unable to send information directly to CHs. All intra-cluster and inter-cluster activities receive resources from each CH. Depending on the cluster's node matter, different amounts of electrical energy are used for intra-cluster activity. With battery-powered network nodes, data aggregation and security are the main research concerns in WSNs. Additionally, information must be sent to the sink safely. To choose an Aggregator Node (AN), the earlier researchers examined a variety of factors, including distance and residual energy. Aggregator nodes use energy more quickly than other nodes and are more sensitive to intrusions. Collaboratively choosing an AN is crucial to overcoming this. In order to maximize network lifetime, secure data transmission, and network throughput, routing protocols must be designed and put into place before choosing an AN. 2. Related Work Although information integration is a commonly used technique in wireless sensor networks, it is not an easy task; many factors affect sensor and data aggregation performance. There are numerous data aggregation scenarios that reduce congestion, network energy, data accuracy latency, and data aggregation rate. This section addresses the many ways of data aggregation for maximizing the network life time that has been reported in literature. In order to prolong operating duration and preserve system functionality, energy-saving strategies for WSN are essential. By minimizing the frequency of transmission and optimizing data communication, energy-efficient communication protocols can reduce the amount of energy used for data transmission. These protocols must, however, strike a balance between the requirement for dependable data delivery, possible latency increases, and energy savings. Sensors should need a certain amount of energy to collect data and a significant quantity of energy to transmit data. When a data aggregation strategy maximizes functionality while consuming the least amount of energy in WSNs, it is considered energy efficient. As more data aggregation cycles are completed till the initial sensor node holds energy, the network lifetime increases due to lower energy consumption. There are numerous data aggregation approaches are used in the arena of wireless sensor networks. Using Adaptive aggregation approach (ADA) sensor and cluster node load is transformed to base station [ 8 ], shows spatial-temporal correlation in terms of two-dimensional characteristics. The base station determines the sensor nodes reporting frequency and cluster nodes aggregating frequency. Distributed algorithm ensures two or more nodes transmit data simultaneously without any interference achieved by orchestrating data aggregation of transmission lines with different time periods without overlapping [ 9 ]. Data delivery ratio is minimized with the help of greedy approach. Delay in data delivery ratio of greedy schedule algorithm is 24D + 6 Delta + 16, which represents a significant improvement over the previous algorithm's latency bound of (Delta - Delta 1) R. An adaptive strategy is included in the algorithm to periodically update agenda dynamically when there is failure of nodes fail or additional nodes scale the network. Decision process model [ 10 ] helps nodes determine optimal times for forwarding samples, balancing energy consumption and delay in communication. When certain statistical conditions are met regarding sample arrival and channel availability, optimal control-limit policies can be implemented effectively; otherwise, learning algorithms can be used to approximate solutions. DEEG protocol allows nodes to independently decide to become cluster heads based on their energy levels and signal strength, enhancing energy efficiency in data aggregation [ 11 ]. DEEG demonstrates a significant improvement in sensor network lifetime, achieving up to 1800% longer lifespan compared to the LEACH protocol and 300% longer than PEGASIS. Consistency sensed information in the DEEG protocol is far better than that in LEACH and PEGASIS, as all nodes in the network remain operational until the preceding 40 rounds before final node perishes. Distributed compressive sparse sampling (DCSS) algorithm allows recovery of n-dimensional information by querying a small number of sensors (m ≪ n), significantly reducing the number of required sensor readings. The use of a sparse binary measurement matrix, designed with an unbalanced expander graph, enhances the performance of compressed sensing schemes by improving recovery accuracy while minimizing the number of sensors needed [ 12 ]. The DCSS algorithm demonstrates robustness against unwanted random error measurements and does not depend on regular sensor deployments, which contributes to lower in-network communication costs and reduced computational demands. Data aggregation in wireless sensor networks reduces resource consumption by combining data and eliminating redundancy. Most existing solutions for data aggregation are static and do not adapt to evolving network constraints like energy, bandwidth, overhead, and transmission delay. Feedback control system depicted in [ 13 ] dynamically optimize data aggregation based on environmental behaviour and application requirements, although it remains a relatively new area of research. Integrity of aggregation results in wireless sensor networks by allowing the base station to verify results immediately upon receipt was illustrated in EIPDAP [ 14 ]. Additional integrity check phase eliminated in EIPDAP results in minimized energy consumption and network delay, which is common in existing protocols. However, protocol achieves high optimal communication overhead per node, represented as O(Δ), where Δ is represented as degree of aggregation tree, making it highly efficient compared to other schemes. Data aggregation routing in wireless sensor networks (WSN) combines several data packets received from the sensor nodes to reduce energy depletion and number of transmissions [ 15 ]. Primary data aggregating strategies used in are mobile agent and sever which depicts client server model of networks. The routing protocols for data aggregation are classified based on network architecture and routing models, addressing key issues in both models. A noble routing protocol illustrated [ 16 ] is derived from LEACH protocol enhances energy efficiency in wireless sensor networks (WSN) by reducing the amount of data transmitted to the base station, which is a major source of energy dissipation. A nodes adaptive schedule is designed to minimize data transmission by addressing the overlap of detection regions. The data aggregation process is integrated into packet transmission, allowing for more effective data management compared to traditional LEACH protocols. Aggregator Nodes (ANs) in wireless sensor networks helps to decrease the number of packets sent, which reduces energy consumption during communication. Distributed Data Aggregation Protocol (DDAP) [ 17 ] is a self-organizing method that employs randomly chosen ANs to optimize data gathering. Combined with DDAP, the Geographical Routing with Aggregation Nodes (GRAN) protocol, which builds upon the GOAFR routing algorithm, substantially decreases data traffic and extends longevity of networks. To improve data secrecy and save power, a distributed aggregation algorithm employing homomorphic trapdoor permutation is recommended. Peer verification is a feature of the protocol that involves at least k peer nodes validating data and sender nodes. A fully distributed approach is made possible by the suggested protocol [ 18 ] for safe data aggregation, which uses a two hop verification mechanism to guarantee data integrity without requiring a central base station. According to simulation results, the new protocol solves the problems brought on by the special characteristics of aggregated data while simultaneously saving energy and preserving data integrity. SAOP suggested architecture acts as a middleware component, making it easier to aggregate data in restricted resource environments a prevalent problem in WSNs. The study highlights SOAP's [ 19 ] viability in resource-constrained environments by comparing its efficacy for data aggregation in WSNs to traditional networks. Instead of using conventional 2-hop routing, the enhanced LEACH protocol utilizes multi-hop routing to increase data aggregation in wireless sensor networks. Energy efficiency is the main focus of this new protocol, which is vital to extending the nodes' lifespan in wireless sensor networks. According to the simulation results, the enhanced LEACH protocol performs better than the traditional LEACH in terms of energy efficiency when acquiring and combining data. With minimal communication overhead, a secure data aggregation technique [ 20 ] for wireless sensor networks (WSNs) ensures fundamental security requirements such source authentication, data secrecy, and data consistency. The protocol uses a message authentication code (MAC) that guarantees data integrity and authenticity and symmetric encryption to safeguard data secrecy. To find and stop false or corrupted data from influencing the final aggregated results, an anomaly detection algorithm is incorporated. Number of cluster heads needed, the application's uninterrupted coverage ratio, residual energy, and node degree are all taken into consideration while identifying cluster heads in CDAT [ 21 ]. By efficiently controlling energy consumption while preserving the intended quality of service (QoS), CDAT outperforms other protocols like LEACH and PEGASIS in terms of network reliability. Effective routing is ensured using EML-DA, [ 22 ] demonstrates a hybrid technique for Cluster Head (CH) selection and robust data aggregation. Artificial Neural Networks (ANN) are used for optimal CH selection based on variables including bandwidth, distance, and residual energy. At the CH node, data aggregation is accomplished through Independent Component Analysis (ICA), which efficiently minimizes redundant data and lowers energy usage through differential entropy and computational efficiency. Neural networks have been employed [ 23 ] to precisely categorize and process sensor data while minimizing errors and noise, deep learning techniques more specifically the ACNM method improve data aggregation. The efficiency of data delivery to its destination is enhanced by the ACNM protocol, which guarantees that aggregated data is transferred with the least amount of delay. By learning from the data, identifying errors, and enhancing the data prior to aggregation, machine learning plays a critical part in this process. Multi-hop information-centric strategy [ 24 ] minimizes information loss while significantly reducing data traffic in information-centric networks. Large-scale IoT and sensor applications are more efficient when data aggregation techniques are incorporated into the ICN paradigm. Seafloor imagery from autonomous synthetic aperture sonar (SAS) systems [ 25 ] can be classified using unsupervised feature learning made possible by deep learning algorithms, especially those that use auto encoders in acoustic sensors. Complex seabed features in SAS photos can be better understood and classified by using generative models in deep architectures. Employing a generator and a discriminator neural network, deep learning models [ 26 ] more especially, generative adversarial networks, or GANs—are used to create new objects or images. This approach easily arranges neural layers to reduce errors, enabling high-quality machine learning outcomes that were previously only possible with specialized software tools. LTDR technique [ 27 ] optimizes work allocation in large-scale data processing by combining deep learning and reinforcement learning with a fat-tree structure. To enhance node mapping decision-making, a virtual network mapping approach based on Q-learning and deep convolutional neural networks is used. This approach satisfies task needs in big data contexts while greatly increasing long-term income and physical resource utilization. Multi-hop reception is usually required to convey this particular information to a central gathering point. Generally speaking, it might currently be examined and improved for further use. It is desirable to reduce interaction as it is expensive in terms of energy usage and can improve the quality of life in the community. This is a crucial feature of a WSN. Analysing extensive data and identifying its characteristics within a wireless sensor network mirrors the function of neural network based data aggregation. Consequently, neural networks can be effectively employed for data aggregation in these networks. By creating a neural network model for each cluster, the cluster head can aggregate data from its associates, derive a representative feature vector for the cluster, and transmit this concise vector to the aggregation node. This approach significantly reduces communication overhead, improves data communication efficiency, lowers energy depletion, and ultimately prolongs the network's operational lifespan. While current data aggregation algorithms optimize for energy, bandwidth, and communication, they fail to account for the inherent inaccuracy of real-world sensor data. Sensor nodes are susceptible to producing erroneous information, including false readings from intruders and environmental noise. This paper evaluates the operational efficiency and energy consumption of different data aggregation approaches. We also introduce a porpoise algorithm optimized, deep learning data aggregation strategy intended for mobile heterogeneous and homogenous wireless sensor networks, which uses weighted clustering to enhance data accuracy. To improve data aggregation accuracy and minimize unnecessary data transfer in WSNs, we propose a technique that retrieves original data features, performs aggregation, and leverages deep learning to optimize the neural network. Cluster head at that moment forwards sophisticated data to sink, facilitating efficient intra network data aggregation. 3. Proposed Methodology Deep learning-based aggregation methods can substantially enhance the performance of WSNs by boosting security, data throughput, and energy efficiency, overcoming many of their inherent limitations [ 28 ]. Deep learning-based data integration algorithm designed to maximize clustering routine in wireless sensor networks, targeting enhanced energy efficiency along with extended network lifetime. Wireless sensor Network generates a enormous volume of information to be transmitted over a limited bandwidth resources. Deep learning network based stacked auto encoder, was employed to learn a lower-dimensional latent representation of sensor readings. SAE facilitates the extraction of salient features from raw sensory information, which is used for tasks like classification and prediction. Cluster head selection is performed using metrics including node density, residual energy, and cluster size. The sink node trains a deep learning model to derive and broadcast optimal parameters, effectively reducing redundant data transmission and minimizing energy expenditure. Porpoise Optimization attains high energy efficiency by means of three stages, Cluster head selection, optimal route selection and hidden layer processing. 3.1 Stacked Auto Encoder Model WSNs in military applications are generally recognized as a highly collaborative technology across the entire battlefield. Stacked auto encoders present several benefits over conventional neural networks, including enhanced generalization performance. Their non-linear transformation, achieved through the composition of stacked hidden layer mappings, is a key factor. Notably, the algorithm's structure allows for easy modification of the hidden layer mapping functions, offering flexibility without requiring fundamental changes to the algorithm itself, and with a relatively straightforward setup involving few parameters. To effectively extract most important features while discarding irrelevant information, encoder alters high dimensional input data to a compact, lower dimensional latent space representation. Unlike raw sensor data, this compressed latent space requires less bandwidth for transmission by leveraging a Gaussian distribution described by a probability density function (PDF), mean, and standard deviation. Latent space captures underlying patterns, trends, and correlations that may not be immediately visible in the original data. Architecture and functionality of stacked autoencoders are rooted in the principles of autoencoders. Autoencoders provide the essential framework, both in terms of design and operation, for constructing stacked autoencoders. A Stacked Auto encoder, depicted in Fig. 1 , serves as the model within this deep learning framework. The Stacked Auto encoder (SAE), a deep neural network architecture Fig. 1 , is particularly effective for unsupervised learning. It comprises multiple encoders (E1, E2, ..., E N ) and corresponding decoders (D1, D2, ..., D N ). A Symmetric SAE is employed when the encoder and decoder layers maintain equal dimensionality. The bottleneck layer, denoted as B, represents the latent space, which encapsulates the essential data features. The architecture includes four layers: a response layer provides input, a yield layer provides the sophisticated output, and two middle hidden layers. These middle layers use the Rectified Linear Unit (ReLU) activation function encodes the input received from initial layer. The encoding function of staked auto encoder using ReLU activation function can be written as in Eq. (1) M i = ReLU(W i *m i−1 +b i ) (1) m i−1 ​ is the output of aforementioned layer or input data, W i ​ and b i ​ are the weight and bias for layer i, m i ​ is the encoded representation at the i th layer. Auto encoders learn in a self-supervised manner because their training objective is to reconstruct the input itself. The training process continually works to decrease error between input response and resulting output. Mean square error cost function through ReLU activation function is given in Eq. ( 2 ) $$\:Err=\frac{1}{N}\sum\:_{i=1}^{N}{(Xi-Yi)}^{2}$$ 2 To efficiently control the number of parameters in the network, a sparsity term is introduced. This additional factor effectively regulates the extent of neuron activations. Specifically, A neuron is deemed to be active if its output is greater than or equal to 1, and inactive if its output is close to zero. The gradients of the ReLU activation function is from 0 to ∞ Outputs from the hidden layers are fed as input to softmax classifier. Target classes (N) from the middle layers are classified using the softmax classifier. ReLU activation function accomplishes the encoding in the hidden layers. Softmax classifier then generates the output data Oi​, as described in the following Eq. ( 3 ). $$\:{O}_{i}=\frac{\text{e}\text{x}\text{p}({W}_{i}m+{b}_{i})}{\sum\:_{j=1}^{N}\text{e}\text{x}\text{p}({W}_{j}m+{b}_{j})}$$ 3 Where m is the output of the middle layers, W i ​ and b i ​ are the weight and bias for layer i, and N be the target classes. Softmax classifiers are effective in tasks such as fault detection and anomaly detection in sensor environments, as they yield high metrics for correctness, precision, and recall. Clusters are formed by assessing the energy levels of nodes in relation to node density, location, and distance between the nodes, using the Cluster Score (C Score) as a metric. Nodes with related characteristics are grouped into distinct clusters. Cluster head is then selected through a clustering process designed to enable effective data transmission with minimal delay using Cluster Head Score (CH Score) as a metric. Vital component of data integration in sensor networks is cluster head selection, which is usually carried by utilizing latent spaces in stacked auto encoders (SAE). By taking into account variables like node energy, node density, location, and cluster size, latent spaces use heuristic functions to maximize communication and increase network longevity. These latent spaces minimize noisy data and resolve nonlinear correlations by utilizing low-dimensional sensor properties produced by unsupervised neural networks. In order to facilitate effective data transfer from the source node to the base station via the best possible path, the Latent Spaces algorithm chooses the cluster head depending on energy levels. The Porpoise method is used to accomplish this ideal multi-path routing, guaranteeing safe data transfer. The en-route maintenance mechanism is used to dynamically select a different route in the case of a route failure. In the early stage, a multi-hop routing protocol exploits a latent space algorithm to cluster sensor nodes. Clustering is primarily based on the nodes' energy levels, a scheme designed to extend the overall network lifespan. Latent space algorithm functions as a vector quantization algorithm splits large collection of sensor nodes into smaller groups, confirming that each group contains a similar number of nodes that are close to one another. The algorithm identifies these nodes based on a cluster score, which measures the similarity between a node's energy level and its nominated cluster head. Initially, all sensor nodes in the network are categorized by their energy, node density, location, and the distances between them. At the time of deployment, all nodes in a Wireless Sensor Network (WSN) begin with the same energy levels. Over a period, each node consumes a portion of this energy for essential operations such as sensing, processing, and communication. Initial energy of the sensor nodes degrades after sensing and monitoring, and the remaining energy is then measured. Eq. ( 4 ) uses the similarity between each cluster head and the residual energy of each node to generate the CH score. $$\:CH\left(i\right)=\:{\vartheta\:}_{1}\left(\frac{1}{{D}_{i}}\right)+{\vartheta\:}_{2}E\left(i\right)+{\vartheta\:}_{3}N\left(i\right)$$ 4 Where \(\:\:{\vartheta\:}_{1}\) , \(\:{\vartheta\:}_{2}\) , \(\:{\vartheta\:}_{3\:}\) are the weights associated with distance among the nodes, Energy and Node density. \(\:\:{D}_{i}\) distance measured between cluster head and sensor node, \(\:E\left(i\right)\) is energy of the sensor node and \(\:N\left(i\right)\) is Node density. Clusters are created based on energy levels, node density, and node distance, and nodes inside each cluster are assessed against the cluster head using the CH Score. Nodes with similar CH scores are grouped to form multiple clusters. A cluster head is chosen during clustering stage to ensure efficient transfer of information with low latency. The proposed strategy gives attention to choosing a cluster head from among the cluster's most energy efficient nodes. 3.2 Stacked Auto Encoders Optimized by Porpoise optimization Combination of multiple encoding and decoding layers in stacked auto encoder enhances learning performance by hierarchically extract the features. Traditional stacked auto encoders frequently rely on random of weights and bias initialization, which remain constant or only partially optimized during the early phases of training, leads to a large number of inefficient nodes that make a negligible contribution to the overall reduction of the cost function. Due to the partial optimization stacked auto encoders would increase the complexity and reduces the generalization ability by incurring large number of hidden layer neurons to achieve adequate performance. Intelligent porpoise optimization technique is renowned for the global search capabilities. To overcome the limitations of traditional stacked auto encoders, this paper proposes a novel model that exploits the Porpoise Optimization Algorithm to optimize the initialization of input weights and thresholds within the stacked auto encoder architecture. The model successfully determines ideal values for the network's starting parameters by utilizing POA's global optimization capabilities, which results in the construction of a more precise and effective network. This optimization framework used to fine-tune the weights and thresholds of stacked auto encoders for cluster head election and data integration in sensor networks. The proposed scheme offers several benefits in large scale sensor network environments. First, Porpoise Optimization Algorithm, existence of global search technique, efficiently involve in local optima also. Second, POA method focuses on improving initial limitations of stacked auto encoder, allowing remaining parameters to be learned more efficiently during training, in contrast to conventional neural networks that necessitate intensive training and optimization across all layers. As a result, this method advances training proficiency and enhances generalization performance. Multi path routing problem in wireless sensor networks is addressed in this paper proposed an enhanced stacked auto encoder technique that is optimized using the Porpoise Optimization technique. Important elements like data transmission distance, energy balance in the network, sensor node remaining energy, and more are all thoroughly taken into account by the suggested algorithm. By doing this, network's coverage is improved, its transmission capacity is increased, its overall energy consumption is decreased, its node energy distribution is balanced, and its operational lifetime is prolonged. The suggested scheme not only exhibits excellent generalization and flexibility, but it also greatly raises the wireless sensor network's overall effectiveness and performance. 3.3 Porpoise Optimized Multi Path Routing Porpoise optimization algorithm pretends scavenging behaviour of porpoises, which uses sonar echo-location and bubble-net hunting patterns. These behaviours are exhibited through dynamic coefficients that control encircling, spiral attack, and exploratory swimming, analogous to exploration exploitation balancing in metaheuristics. Wireless Sensor Networks (WSNs) operate without centralized management or fixed infrastructure and rely on broadcast communication channels. As a result, they lack tamper resistance and are highly susceptible to various security threats. Attackers can easily eavesdrop on network traffic, inject malicious data, replay old messages, or compromise individual sensor nodes. Among the key security challenges, node authentication and privacy protection stand out as major concerns. In this research, we submit a novel Porpoise Optimization Algorithm aimed at efficiently protecting nodes in accordance with security requirements. This algorithm leverages swarm intelligence to automatically identify optimal solutions for node failures, compromised nodes and security analysis. Once the cluster head is selected the CH selects optimum path for communication from source to destination. Whereas the stacked auto encoder routing uses the multipath routing technique for data transmission, diminishes energy depletion and increases network lifespan. SAE selects the optimal route from hop count metrics used for cluster heads selection, selects least hop count from source node to the destination for data transmission. In SAE optimum route from source to destination is calculated by the path score pi based on the residual energy, distance between source to destination and packet delivery ratio along with the threshold values as shown in Eq. ( 5 ). $$\:{P}_{i}=\:\:{\vartheta\:}_{1}\left(\frac{1}{{D}_{i}}\right)+{\vartheta\:}_{2}E\left(i\right)+{\vartheta\:}_{3}{R}_{i}$$ 5 Where \(\:{D}_{i}\) is the distance between source node to destination, \(\:E\left(i\right)\) is energy of node and \(\:{R}_{i}\) is packet delivery ratio. Porpoise optimization algorithm is an iterative optimization algorithm to find the best solution which wold be more efficient and reliable in terms of locally and globally. The detailed architecture of optimization algorithm is depicted in Fig. 1 . There are two major steps in porpoise optimization one is encircling and the other is spiral bubble net. The initial route is carried out by the path score as mentioned in the above equation. Encircling stage is an exploration mechanism that the current route is iteratively adjusted to move closer to the current best routing solution to find the optimal route. Let P(t) be the optimal path at iteration t and \(\:{X}_{i}\left(t\right)\) be the position vector of optimal path P(t) for each sensor nodes. Each routing Paths doesn’t just randomly select instead it is carried out by calculating the packet delivery ratio, residual energy and minimum distance between the source and destination nodes. In encircling stage less optimal routing adjust the parameters by increasing packet delivery ratio and reduced energy consumption in order to prioritize in the next hop that brings the characteristics close to P(t). Let \(\:X\left(t\right)\) be the best solution found so far, distance in path space between the current path and best known solution and it is provided in Eq. ( 6 ) $$\:{D}_{i}\left(t\right)=|C*X\left(t\right)-\:{X}_{i}\left(t\right)|$$ 6 Where C is a constant for randomization factor, for each iteration t + 1 and exploration factor A, the update rule in Eq. ( 7 ) is defined as $$\:{X}_{i}\left(t+1\right)=X\left(t\right)-A*{D}_{i}\left(t\right)$$ 7 Where the exploration factor linearly decreases over time with respect to the controls over convergence. Spiral bubble net provides an exploitative strategy to refine local search around the best optimal route from source to base station to reduce energy depletion, equally divides the load distribution in terms of data integration and provide a reliable communication establishment between source and destination. The ith update of the spiral bubble is provided in Eq. ( 8 ) $$\:{X}_{i}\left(t+1\right)=X\left(t\right)+{D}_{i}\left(t\right).{e}^{al}.Cos\left(2\varPi\:l\right)$$ 8 Where 𝑎 denotes the logarithmic narrowing constant, while 𝑙 refers to a random scalar value within the range [− 1,1]. To enable self-motivated swapping between exploration and exploitation phases, Eq. 9 provide a position routing update mechanism is directed by the following probabilistic rule $$\:{X}_{i}\left(t+1\right)=\left\{\begin{array}{c}X\left(t\right)-A*{D}_{i}\left(t\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:if\:p<0.5\\\:X\left(t\right)+{D}_{i}\left(t\right).{e}^{al}.Cos\left(2\varPi\:l\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:if\:p\ge\:0.5\end{array}\right.$$ 9 Where p € [0,1] Once the optimal solution is produced, the source node communicates the aggregated data to base station. Porpoise algorithm out smart among heuristic intelligent algorithms owing to its frequency tuning capability, enabling automatic changeovers between global and local search. This dynamic control over search correlation, combined simple framework, quick scaling, parallel processing capabilities, ease of use, and great parallel stability, makes it a highly efficient solution. Data integration in a POA-SAE-based wireless sensor network proceeds as follows: First, all network components, including common, routing, and cluster head nodes, are initialized to determine their current states. Next, the monitoring region's clustering structure is established using the sensing node's cluster score. A cluster head node is then randomly selected from each cluster based on its CH Score, and it collects data from its cluster members. Once the WSN clustering stabilizes, the cluster head creates a routing table for shared nodes and communicates member information to the receiving node. Once optimal route is established, the WSN's data integration model undergoes training. This process determines the amount of nodes, weightages, and threshold parameters for its middle learning network. To conserve limited energy of shared sensor nodes and extend the network's lifespan, this SAE-based data integration model training is exclusively performed at receiving node. Receiving node first builds the SAE's network structure using the received information. It then trains this network by matching samples from the cluster head sample database with associate node information. After training, the Sink node obtains the data integration model's parameters. Subsequently, receiving node sends these SAE model network parameters to the relevant cluster head. Each cluster head then uses this trained data integration model to fuse data from its member nodes, extracting features and removing redundant or useless information. The merged data is temporarily stored before being sent back to receiving node. Algorithm proposed in this paper, leverages cluster head node feature extraction level and the softmax classification level to handle feature extraction, data integration, and classification. Processed information is then forward to the receiving node. Based on the Porpoise Optimization Algorithm Stacked Auto encoder (POA-SAE) algorithm, wireless sensor networks follow these multipath routing steps: First, an optimized data integration model is built using stacked auto encoders. To execute multi path routing, this integration model is subsequently incorporated into dynamic heterogeneous wireless sensor network's clustering framework. Enhancing information integration accuracy, lowering data transmission volume, and extending network lifetime are the objectives. 4. Performance Evaluation This study introduces the POA-SAE method, which uses NS3 Simulator for experimental and performance comparison analysis is done through MATLAB to reflect the performance of data integration and multipath routing. assuming that there are enough nodes deployed to ensure a directed node distribution in the monitored area and that a large-scale random node deployment is carried out initially. With 300 sensors, the sensor nodes are set up in a 200x200 m 2 two-dimensional area. There are 50 simulations run in total, with 10 data packets sent from the source to the destination. With an initial node energy of 5J, each packet has a 4 kb capacity. The sensor nodes initialization consumption can reach an average of 0.5 J per node. The porpoise optimization mathematical model provides composite WSN witness function, experiment design and formula metrics and statistical analysis method. Porpoise Optimization Algorithm (POA) is a recent bio-inspired method which models alternating exploration and exploitation behaviours. The Performance of POA-SAE is compared with other benchmark algorithms. The models for a specific network size were simulated 50 times for each case. The results were collected for the clustering time, total energy consumption, latency, Packet delivery ratio, and the remaining energy in the nodes. Each Sensor node is considered as the porpoise which represent the candidate solution in the problem space. Eq. 10 denotes search agent of each sensor node. $$\:{X}_{i}^{t}=\left[{X}_{i,1}^{t},{X}_{i,2}^{t},{X}_{i,3}^{t},\dots\:\dots\:{X}_{i,D}^{t},\right]$$ 10 N denotes number of sensor nodes with value of i varies from i = 1,2,3… N, D dimension of the optimization problem and t represent the iteration index. The initial population of the sensor nodes are generated randomly within the spiral bubble of the search limits and is denoted in Eq. 11 $$\:{x}_{i,d}^{0}=U({L}_{d},{U}_{d})$$ 11 Where \(\:{L}_{d},{U}_{d}\) are the lower and upper bounds with dimension d Default best solution is initialized as represented in Eq. 12 $$\:{X}_{B}^{0}=\text{arg}\text{min}f\left({X}_{i}^{0}\right)$$ 12 After generating new candidates, the fitness is evaluated to provide a best solution. The \(\:{i}^{th}\) fitness is evaluated in Eq. 13 . $$\:{F}_{i}^{t+1}=f\left({X}_{i}^{t+1}\right)$$ 13 If \(\:{F}_{i}^{t+1}<\:{F}_{i}^{t}\) then the new position is updated using the Eq. 14 otherwise the old position is retained, the global best is updated. $$\:{X}_{B}^{t+1}=\text{arg}\text{min}{F}_{i}^{t+1}$$ 14 4.1Simulation Results Convergence plots the best objective value with respect to number of iterations or rounds. It shows how quickly the optimization algorithm finds the best solution. Eq. 15 provide the convergence of POA-SAE algorithm. $$\:F=\:{w}_{1}.{E}_{R}+{w}_{2}.{R}_{i}+{w}_{3}.\frac{1}{L}+{w}_{4}.\frac{1}{{C}_{t}}$$ 15 Where \(\:{w}_{1},\:{w}_{2},{w}_{3},{w}_{4}\) are weights to ensure the balance The convergence behaviour of several algorithms is shown in Fig. 3 . POA-SAE shows a better capacity to identify optimal solutions. POA-SAE exhibits fast convergence; by iterations 20–25, its fitness curve had risen significantly and stabilized close to the global optimum. Its balanced exploration and exploitation strategies, which enable the algorithm avoid becoming stranded in local minima like a porpoise's encircling and bubble-net tactics, are responsible for this effective convergence. Other algorithms, in contrast, show distinct patterns of convergence. Particle Swarm Optimization (PSO) flattens its fitness curve at a lower value than POA's because it converges faster than Genetic Algorithm (GA) but frequently stalls in local optima. Because GA's crossover and mutation activities are inherently random, it exhibits a slower, more gradual convergence. The Low-Energy Adaptive Clustering Hierarchy (LEACH) methodology has a flat, constant performance curve with no true optimization as a baseline. While GA and PSO climb at a considerably slower rate, POA-SAE's fitness increases significantly during the first Phase, or iterations 0–10, as it rapidly improves its cluster-head and route selection. POA-SAE improves its solutions and stabilizes its fitness close to the ideal value during the second phase iterations, which range from 10 to 30. PSO might prematurely plateau during this period, while GA keeps getting better, albeit more slowly. For the Final Phase iterations from 30 to 50 POA-SAE's fitness curve remains flat at a high, stable value. GA's fitness continues to improve but ultimately settles at a lower value than POA, while LEACH remains constant throughout the simulation. 4.2 Residual Energy Residual energy represents the remaining battery power of sensor nodes after each communication round i.e. sensing, transmitting, processing and receiving. Figure 4 illustrates the residual energy trend versus total number of rounds. Average residual energy of sensor node relating to rounds is calculated using the Eq. 16 $$\:{E}_{R}^{i}\left(t\right)={E}_{0}^{i}-\sum\:_{t=1}^{n}\left({E}_{TX}\left(p,d\right)+{E}_{RX}\left(k\right)+{E}_{S}+{E}_{p}\right)$$ 16 Where \(\:{E}_{0}^{i}\) is the initial node energy i.e. 5J, \(\:{E}_{TX,},\:{E}_{RX},\:{E}_{S},\:\&\:{E}_{p}\) are the consumption of transmission, receiving, sensing and processing with respect to packet size p and distance d. Based on the above provided data in Fig. 4 , POA-SAE protocol exhibits superior energy efficiency compared to PSO, GA, WOA, and LEACH. The residual energy analysis shows that POA-SAE consistently maintains a higher level of energy throughout the simulation. After 50 rounds, POA nodes retain approximately 3.2 J of energy, whereas PSO and GA nodes drop to about 2.9 J and 2.7 J, respectively. This suggests that POA-SAE effectively selects more energy-efficient cluster-heads and routing paths. This energy conservation is attributed to POA-SAE's balanced approach, which incorporates exploration for diverse cluster-head selection and exploitation to focus on optimal clusters. POA-SAE prolongs the stability period of the network and delays the occurrence of dead nodes, thereby extending network lifetime. Reduced redundancy of sensed information before communicating it to base station, data aggregation lowers transmission latency and energy usage. POA-SAE achieve better cluster head placement and aggregate more efficiently. Graph mentioned in Fig. 5 shows the effectiveness of data aggregation under different optimization algorithms. POA - SAE maintains the lowest aggregation ratio 0.4 to 0.5, indicating higher redundancy removal and reduced transmissions compared to PSO of about 0.55, GA 0.6, and LEACH at 0.75. LEACH has less aggregation when compared to our proposed POA-SAE optimization algorithm. Lower ratios mean that fewer redundant packets reach the base station, thus saving energy and extending network lifetime. The result confirms that POA-SAE not only balances energy consumption but also improves data aggregation efficiency. The clustering time directly influences the stability and responsiveness of a wireless sensor network (WSN). Figure 6 shows the relationship between clustering time and the number of nodes for different optimization algorithms POA-SAE, PSO, GA, and LEACH. The results show that POA- SAE consistently achieves the lowest clustering time compared to PSO, GA, and LEACH. This is due to the effective balance of exploration and exploitation in POA-SAE’s search dynamics, which accelerates the convergence towards optimal CH selection. In contrast, LEACH shows the highest clustering time since it relies on randomized CH election, which introduces higher overhead. PSO and GA demonstrate moderate clustering performance but are outperformed by POA-SAE. These results confirm the suitability of POA-SAE for large-scale deployments where fast reconfiguration is essential. Figure 7 demonstrates that POA-SAE achieves the most balanced cluster distribution with the lowest variance in cluster sizes. PSO and GA produce reasonably distributed clusters, but slight imbalances remain due to their tendency to converge prematurely. LEACH, on the other hand, exhibits the worst cluster distribution, leading to network hotspots and reduced stability. The balanced distribution achieved by POA-SAE can be attributed to its adaptive exploitation mechanism, which prevents over-concentration of nodes around specific CHs. This ensures fair resource allocation and longer network lifetime. Figure 8 shows how latency changes as the number of rounds increases. The lowest latency is attained via POA-SAE, suggesting effective routing and little congestion. The porpoise-inspired search strategy's optimal path construction is responsible for this improvement. PSO and GA, in contrast, exhibit significant latency reductions but struggle with delayed convergence. Because of its uneven cluster sizes and redundant communication channels, LEACH has the highest latency. POA-SAE's lower latency provides quicker data delivery, which makes it ideal for real-time WSN applications. Considering most of assessed evaluations, POA-SAE shows a notable performance improvement over the baseline algorithms PSO, GA, and LEACH. POA SAE's adaptive exploitation mechanism and spiral movement-based search method, which improves both local and global optimization capability, are responsible for this improvement. The outcomes also demonstrate that POA-SAE is a scalable solution since it can continue to provide consistent performance even as the number of nodes rises. As demonstrated in earlier tests, the enhanced cluster distribution and decreased latency also directly lead to a higher packet delivery ratio and residual energy conservation. The proposed method improves data aggregation and energy efficiency in wireless sensor networks for monitoring regions of different scales in simulations. Therefore, it can be concluded that the strategy proposed in this study effectively improves the overall performance of wireless sensor networks. 5. Conclusion Monitoring several environmental features, including temperature, humidity, pressure, lighting, etc., WSN is made up of numerous dispersed wireless sensor nodes. The paper uses a variety of techniques to enhance energy efficiency, node coverage in WSNs, which automatically increases network’s performance and efficiency. The primary research methodology includes data aggregation, energy efficiency analysis, and the application of POA-SAE algorithms to enhance node deployment, which in turn enhances network efficiency and performance. The study's findings demonstrated that POA-SAE significantly impacted optimization. Wireless Sensor Networks (WSNs) are critical for real time monitoring in various domains but face persistent challenges of energy limitation, redundant data transmission, and inefficient routing. To address these, we propose Porpoise Optimization Algorithm (POA) optimized Stacked Autoencoder (SAE)-based multi-hop routing protocol for efficient data integration Network efficiency is significantly enhanced by combining the global search capacity of POA for the best cluster head selection and routing with the synergy of stacked autoencoders for feature extraction and redundancy elimination. The proposed POA-SAE methodology is clearly superior to baseline techniques like LEACH, GA, and PSO, as shown by the simulation results. With nodes maintaining roughly 3.2 J after 50 rounds as opposed to 2.9 J in PSO and 2.7 J in GA, POA-SAE continuously maintained higher residual energy, indicating nearly 15–20% superior energy efficiency. In terms of data aggregation, POA-SAE achieved an average aggregation ratio of 0.4–0.5, substantially lower than PSO (0.55), GA (0.6), and LEACH (0.75), indicating superior redundancy reduction. Latency was also minimized, with POA-SAE demonstrating up to a 30–35% reduction compared to baseline methods. Furthermore, clustering time was the lowest across all tested scenarios, confirming its scalability for large-scale WSNs. Overall, POA-SAE delivers improved energy efficiency, scalability, reliability, and faster convergence, making it a robust protocol for critical applications. Future research can extend this framework by incorporating adaptive security mechanisms and hybrid optimization approaches to further strengthen resilience and adaptability in dynamic WSN environments. Declarations 6. Funding Declaration The Authors did not receive funds, grant or other support for the submitted work. 7.Conflict of Interest We declare that no financial or non-financial interest to disclose. 8. Acknowledgement: None 9.Author Contributions: Ramesh D. Provide the conceptualization, analysis and methodology of the research work. Jaya T. Provide Suggestion and correction related to methodology and analysis in this research work. References Elhoseny M, Abdelaziz A, Hassanien A (2020) Swarm intelligence–based energy efficient clustering with multihop routing protocol for sustainable wireless sensor networks. Int J Distrib Sens Netw 16(9):1–13 Javaid N, Qureshi TN, Khan AH (2013) EDDEEC: Enhanced developed distributed energy efficient clustering for heterogeneous wireless sensor networks, arXiv preprint arXiv:1303.5274 Qureshi TN, Javaid N, Khan AH (202) Optimized Cluster-Based Dynamic Energy-Aware Routing Protocol for Wireless Sensor Networks in Agriculture Precision. Int. J. Commun. Syst., vol. 33, no. 13, pp. 1–16 Reddy K (2021) Hybrid optimization algorithm for security aware cluster head selection process to aid hierarchical routing in wireless sensor networks. IET Commun 15(7):821–832 Md A, Rahman, Matin MA (2020) Energy-efficient routing in wireless sensor networks: A comprehensive survey. Algorithms, vol. 13, no. 3, pp. 1–38 Sunithar S, Rajkumar R, Anitha M (2024) An Energy-Efficient and Secure WSN Routing Protocol Using Bayesian Networks and Elitist Genetic Algorithms. J Electr Syst Automat 57(6):610–619 Ganesh R, Amutha R (2013) Efficient and Secure Routing Protocol for Wireless Sensor Networks through SNR based Dynamic Clustering Mechanisms (ESRPSDC), arXiv preprint arXiv:1306.0312 Chen H, Mineno H, Mizuno T (2008) Adaptive data aggregation scheme in clustered wireless sensor networks. Comput Commun 31(15):3579–3585 Yu B, Li J, Li Y (2009) Distributed Data Aggregation Scheduling in Wireless Sensor Networks. In IEEE INFOCOM 2009. IEEE. https://doi.org/10.1109/infcom.2009.5062140 Ye Z, Abouzeid AA, Ai J (2007) Optimal Policies for Distributed Data Aggregation in Wireless Sensor Networks. In IEEE INFOCOM 2007–26th IEEE International Conference on Computer Communications (pp. 1676–1684). IEEE. https://doi.org/10.1109/infcom.2007.196 LIU M (2005) A Distributed Energy-Efficient Data Gathering and Aggregation Protocol for Wireless Sensor Networks. Journal of Software, vol 16. China Science Publishing & Media Ltd, p 2106. 12 https://doi.org/10.1360/jos162106 Li S, Qi H (2013) Distributed Data Aggregation for Sparse Recovery in Wireless Sensor Networks. In 2013 IEEE International Conference on Distributed Computing in Sensor Systems (pp. 62–69). IEEE. https://doi.org/10.1109/dcoss.2013.64 Boubiche DE, Boubiche S, Bilami A, Toral-Cruz H (2016) Toward Adaptive Data Aggregation Protocols in Wireless Sensor Networks. In Proceedings of the International Conference on Internet of things and Cloud Computing (pp. 1–6). ACM. https://doi.org/10.1145/2896387.2900325 Zhu L, Yang Z, Li M, Liu D (2013) An Efficient Data Aggregation Protocol Concentrated on Data Integrity in Wireless Sensor Networks. In International Journal of Distributed Sensor Networks (Vol. 9, Issue 5, p. 256852). SAGE Publications. https://doi.org/10.1155/2013/256852 Ardakani SP (2017) Data Aggregation Routing Protocols in Wireless Sensor Networks: A Taxonomy. In International journal of Computer Networks & Communications (Vol. 9, Issue 2, pp. 89–107). Academy and Industry Research Collaboration Center (AIRCC). https://doi.org/10.5121/ijcnc.2017.9207 Ji P, Li Y, Jiang J, Wang T (2012) A Clustering Protocol for Data Aggregation in Wireless Sensor Network. In 2012 International Conference on Control Engineering and Communication Technology (pp. 649–652). IEEE. https://doi.org/10.1109/iccect.2012.260 Chatterjee S, Pais AR, Saurabh S (2016) Distributed Data Aggregation in Wireless Sensor Network - with Peer Verification. In Proceedings of the 13th International Joint Conference on e-Business and Telecommunications (pp. 440–445). SCITEPRESS - Science and Technology Publications. https://doi.org/10.5220/0005991604400445 Bagaa M, Lasla N, Ouadjaout A, Challal Y (2007) SEDAN: Secure and Efficient protocol for Data Aggregation in wireless sensor Networks. In 32nd IEEE Conference on Local Computer Networks (LCN 2007) (pp. 1053–1060). IEEE. https://doi.org/10.1109/lcn.2007.81 Al-Yasiri A, Sunley A (2007) Data aggregation in wireless sensor networks using the SOAP protocol. In Journal of Physics: Conference Series (Vol. 76, p. 012039). IOP Publishing. https://doi.org/10.1088/1742-6596/76/1/012039 Bo W, Han-ying H, Wen F (2008) An Improved LEACH Protocol for Data Gathering and Aggregation in Wireless Sensor Networks. In 2008 International Conference on Computer and Electrical Engineering (pp. 398–401). IEEE. https://doi.org/10.1109/iccee.2008.59 YANG J, ZHANG ZHANGD-Y, Y.-Y., WANG Y (2010) Cluster-Based Data Aggregation and Transmission Protocol for Wireless Sensor Networks. Journal of Software, vol 21. China Science Publishing & Media Ltd, pp 1127–1137. 5 https://doi.org/10.3724/sp.j.1001.2010.03534 Sudha C, Suresh D, Nagesh A (2021) An Enhanced Machine learning data aggregation model for Efficient data processing in wireless sensor networks. In 2021 6th International Conference on Communication and Electronics Systems (ICCES) (pp. 1748–1753). IEEE. https://doi.org/10.1109/icces51350.2021.9489124 C S, D, S., A N (2021) Accurate Data Aggregation Created by Neural Network and Data Classification Processed Through Machine Learning in Wireless Sensor Networks. Res Square Platf LLC. https://doi.org/10.21203/rs.3.rs-895195/v1 Pellenz ME, Lachowski R, Jamhour E, Brante G, Moritz GL, Souza RD (2021) In-Network Data Aggregation for Information-Centric WSNs using Unsupervised Machine Learning Techniques. In 2021 IEEE Symposium on Computers and Communications (ISCC) (pp. 1–7). IEEE. https://doi.org/10.1109/iscc53001.2021.9631416 Chen JL, Summers JE (2016) Deep neural networks for learning classification features and generative models from synthetic aperture sonar big data. In Proceedings of Meetings on Acoustics. Acoustical Society of America. https://doi.org/10.1121/2.0000458 Ilichev VY, Chukhraev IV (2021) Data processing using deep learning of the generative-adversarial neural network (GAN). In Neurocomputers. Publishing house Radiotekhnika https://doi.org/10.18127/j19998554-202105-04 Wu C, Xu G, Ding Y, Zhao J (2019) Explore Deep Neural Network and Reinforcement Learning to Large-scale Tasks Processing in Big Data. International Journal of Pattern Recognition and Artificial Intelligence, vol 33. World Scientific Pub Co Pte Ltd, p 1951010. https://doi.org/10.1142/s0218001419510108 Hu T (2023) Data Fusion Algorithm For Wireless Sensor Networks Based On Cluster Routing And Deep Learning. In Journal of Physics: Conference Series (Vol. 2504, Issue 1, p. 012043). IOP Publishing. https://doi.org/10.1088/1742-6596/2504/1/012043 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 14 May, 2026 Reviews received at journal 07 May, 2026 Reviewers agreed at journal 25 Apr, 2026 Reviewers agreed at journal 23 Apr, 2026 Reviewers agreed at journal 23 Apr, 2026 Reviewers invited by journal 20 Apr, 2026 Editor assigned by journal 11 Apr, 2026 Submission checks completed at journal 09 Oct, 2025 First submitted to journal 07 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7799400","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":629628384,"identity":"81c4a31e-a0b5-46d9-be73-68731d792fc4","order_by":0,"name":"Ramesh. D","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIiWNgGAWjYLACxga2+v3nmw8AmRIyxGrhY2y4cSwBpIWHWC1yjA0HcgxAbMJazNtPJz78ucOMmbHhzOdXN2oseBjYDx/dgE+LzJnczca8Z9LYmJl7t1nnHAM6jCct7QY+LRIMudukGduO8bAxnN1mnMMG1CLBY4ZfC//b7T9/tv0HKs15ZpzzjxgtErnbGHjb2AwkGHKYH+e2EaXl7WZpoJYEA4ljZsy5fRI8bAT9wp+78eNPkBb+5sefc77VyfGzHz6GVwsyYJMAk8QqBwHmD6SoHgWjYBSMgpEDAHnCRnKJdlx3AAAAAElFTkSuQmCC","orcid":"","institution":"Anna University, Chennai","correspondingAuthor":true,"prefix":"","firstName":"Ramesh.","middleName":"","lastName":"D","suffix":""},{"id":629628385,"identity":"1592c5c8-dc94-45f3-83ed-975fce66d1d7","order_by":1,"name":"T. Jaya","email":"","orcid":"","institution":"Saveetha Engineering College","correspondingAuthor":false,"prefix":"","firstName":"T.","middleName":"","lastName":"Jaya","suffix":""}],"badges":[],"createdAt":"2025-10-07 12:31:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7799400/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7799400/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108183322,"identity":"cca5e7b7-52fc-4f0f-b0c9-fe600a0e4b06","added_by":"auto","created_at":"2026-04-30 09:00:41","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":46741,"visible":true,"origin":"","legend":"\u003cp\u003eArchitecture of stacked Auto Encoder\u003c/p\u003e","description":"","filename":"image2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/c151adf8a466ef589c6f565a.jpeg"},{"id":108183318,"identity":"58aa7fd3-c034-45d7-ade0-b0556e0f6f1d","added_by":"auto","created_at":"2026-04-30 09:00:40","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":188230,"visible":true,"origin":"","legend":"\u003cp\u003eSAE POA Architecture\u003c/p\u003e","description":"","filename":"image3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/683857dc11630ef1b3c5a6c7.jpeg"},{"id":108185223,"identity":"2c0a305d-d15d-488d-b93d-af7f7add4c5e","added_by":"auto","created_at":"2026-04-30 09:05:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":255827,"visible":true,"origin":"","legend":"\u003cp\u003eConvergence Graph iterations vs Best Objective value\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/6ad3643e142118118b92b8fb.png"},{"id":108183719,"identity":"2d9678bf-f943-4373-a00b-d9f9c40a59b1","added_by":"auto","created_at":"2026-04-30 09:02:33","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":192976,"visible":true,"origin":"","legend":"\u003cp\u003eRemaining Energy VS Number of rounds\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/2f447016ed8adb4002591e03.png"},{"id":108183770,"identity":"8d92b2df-81ec-4000-bf27-91a7da66eb30","added_by":"auto","created_at":"2026-04-30 09:02:43","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":233900,"visible":true,"origin":"","legend":"\u003cp\u003eData aggregation ration Vs Number of rounds\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/fecfc8d268386e2a508a7f50.png"},{"id":108183307,"identity":"5bab2928-1b18-4295-ab38-04cef0da22ae","added_by":"auto","created_at":"2026-04-30 09:00:37","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":155745,"visible":true,"origin":"","legend":"\u003cp\u003eClustering time Vs Number of Nodes\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/ce06e4ac85d2c0b141fed03f.png"},{"id":108184195,"identity":"5c83cbde-2a5f-43d0-a76d-0ede3fa9eadb","added_by":"auto","created_at":"2026-04-30 09:03:34","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":82508,"visible":true,"origin":"","legend":"\u003cp\u003eCluster Distribution\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/f7a5d32edb8942f5ad5cbe60.png"},{"id":108183319,"identity":"647c1672-6a94-4859-9a94-8065e54ad4fb","added_by":"auto","created_at":"2026-04-30 09:00:40","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":143430,"visible":true,"origin":"","legend":"\u003cp\u003eLatency VS Number of Rounds\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/abc67734a134828bf8f8d6f1.png"},{"id":108185391,"identity":"7cc2e6ce-ebaa-4fc6-8a2c-1a2a247df84f","added_by":"auto","created_at":"2026-04-30 09:05:41","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1472716,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7799400/v1/05913a5a-6433-4869-bbe9-0f8a0c265ce6.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Porpoise Optimized Deep Learning Based Multi-Hop Routing Protocol for Efficient Data Integration in WSN","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWireless Sensor networks comprised of multiple interconnected nodes, deployed in areas that are inaccessible to humans without the need for any physical media. Sensor nodes assist other nodes understand and interpret information. Lifetime of the network is indirectly proportional to its energy usage, results in increased lifespan because of its lower energy consumption [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. A system that is only productive for a brief amount of time poses a severe hazard. Techniques for overloading short-lived networks include nap cycle arrangement, information reduction, and stereo recovery [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. All objects in the network can be logically or physically communicated with by community nodes; the application focus on the subsequent social actor actions. The rational function of the links primarily determines the logical architecture. Sometimes it's random, and other times it's a plan. Centralized approaches are suitable as a system's computational energy capabilities rely mostly on a single magnetic generator [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. This specific unit is in charge of managing, coordinating, and processing the sensed information pursuits in such cases. Sensor nodes deployed in a hostile environment, security concerns, energy usage, data secrecy, delay in packet transmission between two sensor nodes, trust worthiness of data, and data aggregation become crucial [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSensor system can achieve emergent behaviour in which nodes communicate independently and coordinate on their own. The objective is to accomplish things that go beyond what each node alone can do. WSN security is used in a diverse situation, including medical, military, civilian duties, and disaster relief. Furthermore, in order to collect data sensor nodes transmission of packets from sensed sensor node to respective Base Station (BS) is essential in WSN. Another important aspect of WSN that could be used to enhance the performance of computing is parallelism [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. A good routing procedure is necessary for the efficient transfer of information, and from the perspective of the consumer, fake or even delayed information is insufficient. Routing protocols are divided into a few groups based on community structure and process technique. Routing protocols can be categorized as location-dependent, Layered, or possibly based on the design of the system. Despite their emphasis on security, all routing processes have pros and cons. To manage the integrity, confidentiality, and authenticity of sensed data, a protective framework is required [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Both efficacy and communal life expectancy were linked to protection; Recently, a lot of research has been done to develop dependable and efficient routing methods. Spoofing or even altering the information path, Sybil encounter, choosy forwarding, wormhole encounter, sinkhole encounter, and HELLO foods encounter are some of the many attacks that occur throughout the system-level process [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. In order to achieve energy effectiveness, clustering systems may also be preferred only in multihop information exchange. To achieve greater energy efficiency and network scalability, sensor nodes are arranged directly into clusters. The BS's cluster heads (CHs) act as routers, sending data from CHs located farther away. This is merely due to the fact that certain nodes, which have a wealth of web resources, are unable to send information directly to CHs. All intra-cluster and inter-cluster activities receive resources from each CH. Depending on the cluster's node matter, different amounts of electrical energy are used for intra-cluster activity.\u003c/p\u003e \u003cp\u003eWith battery-powered network nodes, data aggregation and security are the main research concerns in WSNs. Additionally, information must be sent to the sink safely. To choose an Aggregator Node (AN), the earlier researchers examined a variety of factors, including distance and residual energy. Aggregator nodes use energy more quickly than other nodes and are more sensitive to intrusions. Collaboratively choosing an AN is crucial to overcoming this. In order to maximize network lifetime, secure data transmission, and network throughput, routing protocols must be designed and put into place before choosing an AN.\u003c/p\u003e"},{"header":"2. Related Work","content":"\u003cp\u003eAlthough information integration is a commonly used technique in wireless sensor networks, it is not an easy task; many factors affect sensor and data aggregation performance. There are numerous data aggregation scenarios that reduce congestion, network energy, data accuracy latency, and data aggregation rate. This section addresses the many ways of data aggregation for maximizing the network life time that has been reported in literature. In order to prolong operating duration and preserve system functionality, energy-saving strategies for WSN are essential. By minimizing the frequency of transmission and optimizing data communication, energy-efficient communication protocols can reduce the amount of energy used for data transmission. These protocols must, however, strike a balance between the requirement for dependable data delivery, possible latency increases, and energy savings. Sensors should need a certain amount of energy to collect data and a significant quantity of energy to transmit data. When a data aggregation strategy maximizes functionality while consuming the least amount of energy in WSNs, it is considered energy efficient. As more data aggregation cycles are completed till the initial sensor node holds energy, the network lifetime increases due to lower energy consumption. There are numerous data aggregation approaches are used in the arena of wireless sensor networks. Using Adaptive aggregation approach (ADA) sensor and cluster node load is transformed to base station [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], shows spatial-temporal correlation in terms of two-dimensional characteristics. The base station determines the sensor nodes reporting frequency and cluster nodes aggregating frequency.\u003c/p\u003e \u003cp\u003eDistributed algorithm ensures two or more nodes transmit data simultaneously without any interference achieved by orchestrating data aggregation of transmission lines with different time periods without overlapping [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Data delivery ratio is minimized with the help of greedy approach. Delay in data delivery ratio of greedy schedule algorithm is 24D\u0026thinsp;+\u0026thinsp;6 Delta\u0026thinsp;+\u0026thinsp;16, which represents a significant improvement over the previous algorithm's latency bound of (Delta - Delta 1) R. An adaptive strategy is included in the algorithm to periodically update agenda dynamically when there is failure of nodes fail or additional nodes scale the network. Decision process model [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] helps nodes determine optimal times for forwarding samples, balancing energy consumption and delay in communication. When certain statistical conditions are met regarding sample arrival and channel availability, optimal control-limit policies can be implemented effectively; otherwise, learning algorithms can be used to approximate solutions. DEEG protocol allows nodes to independently decide to become cluster heads based on their energy levels and signal strength, enhancing energy efficiency in data aggregation [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. DEEG demonstrates a significant improvement in sensor network lifetime, achieving up to 1800% longer lifespan compared to the LEACH protocol and 300% longer than PEGASIS. Consistency sensed information in the DEEG protocol is far better than that in LEACH and PEGASIS, as all nodes in the network remain operational until the preceding 40 rounds before final node perishes.\u003c/p\u003e \u003cp\u003eDistributed compressive sparse sampling (DCSS) algorithm allows recovery of n-dimensional information by querying a small number of sensors (m ≪ n), significantly reducing the number of required sensor readings. The use of a sparse binary measurement matrix, designed with an unbalanced expander graph, enhances the performance of compressed sensing schemes by improving recovery accuracy while minimizing the number of sensors needed [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The DCSS algorithm demonstrates robustness against unwanted random error measurements and does not depend on regular sensor deployments, which contributes to lower in-network communication costs and reduced computational demands. Data aggregation in wireless sensor networks reduces resource consumption by combining data and eliminating redundancy. Most existing solutions for data aggregation are static and do not adapt to evolving network constraints like energy, bandwidth, overhead, and transmission delay. Feedback control system depicted in [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] dynamically optimize data aggregation based on environmental behaviour and application requirements, although it remains a relatively new area of research.\u003c/p\u003e \u003cp\u003eIntegrity of aggregation results in wireless sensor networks by allowing the base station to verify results immediately upon receipt was illustrated in EIPDAP [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Additional integrity check phase eliminated in EIPDAP results in minimized energy consumption and network delay, which is common in existing protocols. However, protocol achieves high optimal communication overhead per node, represented as O(Δ), where Δ is represented as degree of aggregation tree, making it highly efficient compared to other schemes. Data aggregation routing in wireless sensor networks (WSN) combines several data packets received from the sensor nodes to reduce energy depletion and number of transmissions [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Primary data aggregating strategies used in are mobile agent and sever which depicts client server model of networks. The routing protocols for data aggregation are classified based on network architecture and routing models, addressing key issues in both models. A noble routing protocol illustrated [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] is derived from LEACH protocol enhances energy efficiency in wireless sensor networks (WSN) by reducing the amount of data transmitted to the base station, which is a major source of energy dissipation. A nodes adaptive schedule is designed to minimize data transmission by addressing the overlap of detection regions. The data aggregation process is integrated into packet transmission, allowing for more effective data management compared to traditional LEACH protocols.\u003c/p\u003e \u003cp\u003eAggregator Nodes (ANs) in wireless sensor networks helps to decrease the number of packets sent, which reduces energy consumption during communication. Distributed Data Aggregation Protocol (DDAP) [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] is a self-organizing method that employs randomly chosen ANs to optimize data gathering. Combined with DDAP, the Geographical Routing with Aggregation Nodes (GRAN) protocol, which builds upon the GOAFR routing algorithm, substantially decreases data traffic and extends longevity of networks. To improve data secrecy and save power, a distributed aggregation algorithm employing homomorphic trapdoor permutation is recommended. Peer verification is a feature of the protocol that involves at least k peer nodes validating data and sender nodes. A fully distributed approach is made possible by the suggested protocol [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] for safe data aggregation, which uses a two hop verification mechanism to guarantee data integrity without requiring a central base station. According to simulation results, the new protocol solves the problems brought on by the special characteristics of aggregated data while simultaneously saving energy and preserving data integrity. SAOP suggested architecture acts as a middleware component, making it easier to aggregate data in restricted resource environments a prevalent problem in WSNs. The study highlights SOAP's [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] viability in resource-constrained environments by comparing its efficacy for data aggregation in WSNs to traditional networks. Instead of using conventional 2-hop routing, the enhanced LEACH protocol utilizes multi-hop routing to increase data aggregation in wireless sensor networks.\u003c/p\u003e \u003cp\u003eEnergy efficiency is the main focus of this new protocol, which is vital to extending the nodes' lifespan in wireless sensor networks. According to the simulation results, the enhanced LEACH protocol performs better than the traditional LEACH in terms of energy efficiency when acquiring and combining data. With minimal communication overhead, a secure data aggregation technique [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] for wireless sensor networks (WSNs) ensures fundamental security requirements such source authentication, data secrecy, and data consistency. The protocol uses a message authentication code (MAC) that guarantees data integrity and authenticity and symmetric encryption to safeguard data secrecy. To find and stop false or corrupted data from influencing the final aggregated results, an anomaly detection algorithm is incorporated. Number of cluster heads needed, the application's uninterrupted coverage ratio, residual energy, and node degree are all taken into consideration while identifying cluster heads in CDAT [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. By efficiently controlling energy consumption while preserving the intended quality of service (QoS), CDAT outperforms other protocols like LEACH and PEGASIS in terms of network reliability. Effective routing is ensured using EML-DA, [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] demonstrates a hybrid technique for Cluster Head (CH) selection and robust data aggregation. Artificial Neural Networks (ANN) are used for optimal CH selection based on variables including bandwidth, distance, and residual energy. At the CH node, data aggregation is accomplished through Independent Component Analysis (ICA), which efficiently minimizes redundant data and lowers energy usage through differential entropy and computational efficiency.\u003c/p\u003e \u003cp\u003eNeural networks have been employed [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] to precisely categorize and process sensor data while minimizing errors and noise, deep learning techniques more specifically the ACNM method improve data aggregation. The efficiency of data delivery to its destination is enhanced by the ACNM protocol, which guarantees that aggregated data is transferred with the least amount of delay. By learning from the data, identifying errors, and enhancing the data prior to aggregation, machine learning plays a critical part in this process. Multi-hop information-centric strategy [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] minimizes information loss while significantly reducing data traffic in information-centric networks. Large-scale IoT and sensor applications are more efficient when data aggregation techniques are incorporated into the ICN paradigm. Seafloor imagery from autonomous synthetic aperture sonar (SAS) systems [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] can be classified using unsupervised feature learning made possible by deep learning algorithms, especially those that use auto encoders in acoustic sensors. Complex seabed features in SAS photos can be better understood and classified by using generative models in deep architectures. Employing a generator and a discriminator neural network, deep learning models [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] more especially, generative adversarial networks, or GANs\u0026mdash;are used to create new objects or images. This approach easily arranges neural layers to reduce errors, enabling high-quality machine learning outcomes that were previously only possible with specialized software tools. LTDR technique [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] optimizes work allocation in large-scale data processing by combining deep learning and reinforcement learning with a fat-tree structure. To enhance node mapping decision-making, a virtual network mapping approach based on Q-learning and deep convolutional neural networks is used. This approach satisfies task needs in big data contexts while greatly increasing long-term income and physical resource utilization.\u003c/p\u003e \u003cp\u003eMulti-hop reception is usually required to convey this particular information to a central gathering point. Generally speaking, it might currently be examined and improved for further use. It is desirable to reduce interaction as it is expensive in terms of energy usage and can improve the quality of life in the community. This is a crucial feature of a WSN.\u003c/p\u003e \u003cp\u003eAnalysing extensive data and identifying its characteristics within a wireless sensor network mirrors the function of neural network based data aggregation. Consequently, neural networks can be effectively employed for data aggregation in these networks. By creating a neural network model for each cluster, the cluster head can aggregate data from its associates, derive a representative feature vector for the cluster, and transmit this concise vector to the aggregation node. This approach significantly reduces communication overhead, improves data communication efficiency, lowers energy depletion, and ultimately prolongs the network's operational lifespan. While current data aggregation algorithms optimize for energy, bandwidth, and communication, they fail to account for the inherent inaccuracy of real-world sensor data. Sensor nodes are susceptible to producing erroneous information, including false readings from intruders and environmental noise. This paper evaluates the operational efficiency and energy consumption of different data aggregation approaches. We also introduce a porpoise algorithm optimized, deep learning data aggregation strategy intended for mobile heterogeneous and homogenous wireless sensor networks, which uses weighted clustering to enhance data accuracy. To improve data aggregation accuracy and minimize unnecessary data transfer in WSNs, we propose a technique that retrieves original data features, performs aggregation, and leverages deep learning to optimize the neural network. Cluster head at that moment forwards sophisticated data to sink, facilitating efficient intra network data aggregation.\u003c/p\u003e"},{"header":"3. Proposed Methodology","content":"\u003cp\u003eDeep learning-based aggregation methods can substantially enhance the performance of WSNs by boosting security, data throughput, and energy efficiency, overcoming many of their inherent limitations [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Deep learning-based data integration algorithm designed to maximize clustering routine in wireless sensor networks, targeting enhanced energy efficiency along with extended network lifetime. Wireless sensor Network generates a enormous volume of information to be transmitted over a limited bandwidth resources. Deep learning network based stacked auto encoder, was employed to learn a lower-dimensional latent representation of sensor readings. SAE facilitates the extraction of salient features from raw sensory information, which is used for tasks like classification and prediction. Cluster head selection is performed using metrics including node density, residual energy, and cluster size. The sink node trains a deep learning model to derive and broadcast optimal parameters, effectively reducing redundant data transmission and minimizing energy expenditure. Porpoise Optimization attains high energy efficiency by means of three stages, Cluster head selection, optimal route selection and hidden layer processing.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Stacked Auto Encoder Model\u003c/h2\u003e \u003cp\u003eWSNs in military applications are generally recognized as a highly collaborative technology across the entire battlefield. Stacked auto encoders present several benefits over conventional neural networks, including enhanced generalization performance. Their non-linear transformation, achieved through the composition of stacked hidden layer mappings, is a key factor. Notably, the algorithm's structure allows for easy modification of the hidden layer mapping functions, offering flexibility without requiring fundamental changes to the algorithm itself, and with a relatively straightforward setup involving few parameters.\u003c/p\u003e \u003cp\u003eTo effectively extract most important features while discarding irrelevant information, encoder alters high dimensional input data to a compact, lower dimensional latent space representation. Unlike raw sensor data, this compressed latent space requires less bandwidth for transmission by leveraging a Gaussian distribution described by a probability density function (PDF), mean, and standard deviation. Latent space captures underlying patterns, trends, and correlations that may not be immediately visible in the original data.\u003c/p\u003e \u003cp\u003eArchitecture and functionality of stacked autoencoders are rooted in the principles of autoencoders. Autoencoders provide the essential framework, both in terms of design and operation, for constructing stacked autoencoders. A Stacked Auto encoder, depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, serves as the model within this deep learning framework.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe Stacked Auto encoder (SAE), a deep neural network architecture Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, is particularly effective for unsupervised learning. It comprises multiple encoders (E1, E2, ..., E\u003csub\u003eN\u003c/sub\u003e) and corresponding decoders (D1, D2, ..., D\u003csub\u003eN\u003c/sub\u003e). A Symmetric SAE is employed when the encoder and decoder layers maintain equal dimensionality. The bottleneck layer, denoted as B, represents the latent space, which encapsulates the essential data features. The architecture includes four layers: a response layer provides input, a yield layer provides the sophisticated output, and two middle hidden layers. These middle layers use the Rectified Linear Unit (ReLU) activation function encodes the input received from initial layer. The encoding function of staked auto encoder using ReLU activation function can be written as in Eq.\u0026nbsp;(1)\u003c/p\u003e \u003cp\u003eM\u003csub\u003ei\u003c/sub\u003e = ReLU(W\u003csub\u003ei\u003c/sub\u003e*m\u003csub\u003ei\u0026minus;1\u003c/sub\u003e+b\u003csub\u003ei\u003c/sub\u003e) (1)\u003c/p\u003e \u003cp\u003em\u003csub\u003ei\u0026minus;1\u003c/sub\u003e​ is the output of aforementioned layer or input data,\u003c/p\u003e \u003cp\u003eW\u003csub\u003ei\u003c/sub\u003e​ and b\u003csub\u003ei\u003c/sub\u003e​ are the weight and bias for layer i,\u003c/p\u003e \u003cp\u003em\u003csub\u003ei\u003c/sub\u003e​ is the encoded representation at the i\u003csup\u003eth\u003c/sup\u003e layer.\u003c/p\u003e \u003cp\u003eAuto encoders learn in a self-supervised manner because their training objective is to reconstruct the input itself. The training process continually works to decrease error between input response and resulting output. Mean square error cost function through ReLU activation function is given in Eq.\u0026nbsp;(\u003cspan refid=\"Equ1\" class=\"InternalRef\"\u003e2\u003c/span\u003e)\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:Err=\\frac{1}{N}\\sum\\:_{i=1}^{N}{(Xi-Yi)}^{2}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eTo efficiently control the number of parameters in the network, a sparsity term is introduced. This additional factor effectively regulates the extent of neuron activations. Specifically, A neuron is deemed to be active if its output is greater than or equal to 1, and inactive if its output is close to zero.\u003c/p\u003e \u003cp\u003eThe gradients of the ReLU activation function is from 0 to \u0026infin;\u003c/p\u003e \u003cp\u003eOutputs from the hidden layers are fed as input to softmax classifier. Target classes (N) from the middle layers are classified using the softmax classifier. ReLU activation function accomplishes the encoding in the hidden layers. Softmax classifier then generates the output data Oi​, as described in the following Eq.\u0026nbsp;(\u003cspan refid=\"Equ2\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:{O}_{i}=\\frac{\\text{e}\\text{x}\\text{p}({W}_{i}m+{b}_{i})}{\\sum\\:_{j=1}^{N}\\text{e}\\text{x}\\text{p}({W}_{j}m+{b}_{j})}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere m is the output of the middle layers, W\u003csub\u003ei\u003c/sub\u003e​ and b\u003csub\u003ei\u003c/sub\u003e​ are the weight and bias for layer i, and N be the target classes. Softmax classifiers are effective in tasks such as fault detection and anomaly detection in sensor environments, as they yield high metrics for correctness, precision, and recall.\u003c/p\u003e \u003cp\u003eClusters are formed by assessing the energy levels of nodes in relation to node density, location, and distance between the nodes, using the Cluster Score (C Score) as a metric. Nodes with related characteristics are grouped into distinct clusters. Cluster head is then selected through a clustering process designed to enable effective data transmission with minimal delay using Cluster Head Score (CH Score) as a metric. Vital component of data integration in sensor networks is cluster head selection, which is usually carried by utilizing latent spaces in stacked auto encoders (SAE). By taking into account variables like node energy, node density, location, and cluster size, latent spaces use heuristic functions to maximize communication and increase network longevity. These latent spaces minimize noisy data and resolve nonlinear correlations by utilizing low-dimensional sensor properties produced by unsupervised neural networks. In order to facilitate effective data transfer from the source node to the base station via the best possible path, the Latent Spaces algorithm chooses the cluster head depending on energy levels. The Porpoise method is used to accomplish this ideal multi-path routing, guaranteeing safe data transfer. The en-route maintenance mechanism is used to dynamically select a different route in the case of a route failure.\u003c/p\u003e \u003cp\u003eIn the early stage, a multi-hop routing protocol exploits a latent space algorithm to cluster sensor nodes. Clustering is primarily based on the nodes' energy levels, a scheme designed to extend the overall network lifespan. Latent space algorithm functions as a vector quantization algorithm splits large collection of sensor nodes into smaller groups, confirming that each group contains a similar number of nodes that are close to one another. The algorithm identifies these nodes based on a cluster score, which measures the similarity between a node's energy level and its nominated cluster head. Initially, all sensor nodes in the network are categorized by their energy, node density, location, and the distances between them. At the time of deployment, all nodes in a Wireless Sensor Network (WSN) begin with the same energy levels. Over a period, each node consumes a portion of this energy for essential operations such as sensing, processing, and communication.\u003c/p\u003e \u003cp\u003eInitial energy of the sensor nodes degrades after sensing and monitoring, and the remaining energy is then measured. Eq.\u0026nbsp;(\u003cspan refid=\"Equ3\" class=\"InternalRef\"\u003e4\u003c/span\u003e) uses the similarity between each cluster head and the residual energy of each node to generate the CH score.\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:CH\\left(i\\right)=\\:{\\vartheta\\:}_{1}\\left(\\frac{1}{{D}_{i}}\\right)+{\\vartheta\\:}_{2}E\\left(i\\right)+{\\vartheta\\:}_{3}N\\left(i\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:{\\vartheta\\:}_{1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\vartheta\\:}_{2}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\vartheta\\:}_{3\\:}\\)\u003c/span\u003e\u003c/span\u003eare the weights associated with distance among the nodes, Energy and Node density.\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:{D}_{i}\\)\u003c/span\u003e\u003c/span\u003e distance measured between cluster head and sensor node, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:E\\left(i\\right)\\)\u003c/span\u003e\u003c/span\u003e is energy of the sensor node and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N\\left(i\\right)\\)\u003c/span\u003e\u003c/span\u003e is Node density.\u003c/p\u003e \u003cp\u003eClusters are created based on energy levels, node density, and node distance, and nodes inside each cluster are assessed against the cluster head using the CH Score. Nodes with similar CH scores are grouped to form multiple clusters. A cluster head is chosen during clustering stage to ensure efficient transfer of information with low latency. The proposed strategy gives attention to choosing a cluster head from among the cluster's most energy efficient nodes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Stacked Auto Encoders Optimized by Porpoise optimization\u003c/h2\u003e \u003cp\u003eCombination of multiple encoding and decoding layers in stacked auto encoder enhances learning performance by hierarchically extract the features. Traditional stacked auto encoders frequently rely on random of weights and bias initialization, which remain constant or only partially optimized during the early phases of training, leads to a large number of inefficient nodes that make a negligible contribution to the overall reduction of the cost function. Due to the partial optimization stacked auto encoders would increase the complexity and reduces the generalization ability by incurring large number of hidden layer neurons to achieve adequate performance.\u003c/p\u003e \u003cp\u003eIntelligent porpoise optimization technique is renowned for the global search capabilities. To overcome the limitations of traditional stacked auto encoders, this paper proposes a novel model that exploits the Porpoise Optimization Algorithm to optimize the initialization of input weights and thresholds within the stacked auto encoder architecture. The model successfully determines ideal values for the network's starting parameters by utilizing POA's global optimization capabilities, which results in the construction of a more precise and effective network.\u003c/p\u003e \u003cp\u003eThis optimization framework used to fine-tune the weights and thresholds of stacked auto encoders for cluster head election and data integration in sensor networks. The proposed scheme offers several benefits in large scale sensor network environments. First, Porpoise Optimization Algorithm, existence of global search technique, efficiently involve in local optima also. Second, POA method focuses on improving initial limitations of stacked auto encoder, allowing remaining parameters to be learned more efficiently during training, in contrast to conventional neural networks that necessitate intensive training and optimization across all layers. As a result, this method advances training proficiency and enhances generalization performance.\u003c/p\u003e \u003cp\u003eMulti path routing problem in wireless sensor networks is addressed in this paper proposed an enhanced stacked auto encoder technique that is optimized using the Porpoise Optimization technique. Important elements like data transmission distance, energy balance in the network, sensor node remaining energy, and more are all thoroughly taken into account by the suggested algorithm. By doing this, network's coverage is improved, its transmission capacity is increased, its overall energy consumption is decreased, its node energy distribution is balanced, and its operational lifetime is prolonged. The suggested scheme not only exhibits excellent generalization and flexibility, but it also greatly raises the wireless sensor network's overall effectiveness and performance.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Porpoise Optimized Multi Path Routing\u003c/h2\u003e \u003cp\u003ePorpoise optimization algorithm pretends scavenging behaviour of porpoises, which uses sonar echo-location and bubble-net hunting patterns. These behaviours are exhibited through dynamic coefficients that control encircling, spiral attack, and exploratory swimming, analogous to exploration exploitation balancing in metaheuristics. Wireless Sensor Networks (WSNs) operate without centralized management or fixed infrastructure and rely on broadcast communication channels.\u003c/p\u003e \u003cp\u003eAs a result, they lack tamper resistance and are highly susceptible to various security threats. Attackers can easily eavesdrop on network traffic, inject malicious data, replay old messages, or compromise individual sensor nodes. Among the key security challenges, node authentication and privacy protection stand out as major concerns. In this research, we submit a novel Porpoise Optimization Algorithm aimed at efficiently protecting nodes in accordance with security requirements. This algorithm leverages swarm intelligence to automatically identify optimal solutions for node failures, compromised nodes and security analysis.\u003c/p\u003e \u003cp\u003eOnce the cluster head is selected the CH selects optimum path for communication from source to destination. Whereas the stacked auto encoder routing uses the multipath routing technique for data transmission, diminishes energy depletion and increases network lifespan. SAE selects the optimal route from hop count metrics used for cluster heads selection, selects least hop count from source node to the destination for data transmission. In SAE optimum route from source to destination is calculated by the path score pi based on the residual energy, distance between source to destination and packet delivery ratio along with the threshold values as shown in Eq.\u0026nbsp;(\u003cspan refid=\"Equ4\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$\\:{P}_{i}=\\:\\:{\\vartheta\\:}_{1}\\left(\\frac{1}{{D}_{i}}\\right)+{\\vartheta\\:}_{2}E\\left(i\\right)+{\\vartheta\\:}_{3}{R}_{i}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e5\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{D}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the distance between source node to destination, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:E\\left(i\\right)\\)\u003c/span\u003e\u003c/span\u003e is energy of node and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{R}_{i}\\)\u003c/span\u003e\u003c/span\u003e is packet delivery ratio.\u003c/p\u003e \u003cp\u003ePorpoise optimization algorithm is an iterative optimization algorithm to find the best solution which wold be more efficient and reliable in terms of locally and globally. The detailed architecture of optimization algorithm is depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. There are two major steps in porpoise optimization one is encircling and the other is spiral bubble net. The initial route is carried out by the path score as mentioned in the above equation. Encircling stage is an exploration mechanism that the current route is iteratively adjusted to move closer to the current best routing solution to find the optimal route. Let P(t) be the optimal path at iteration t and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{X}_{i}\\left(t\\right)\\)\u003c/span\u003e\u003c/span\u003e be the position vector of optimal path P(t) for each sensor nodes. Each routing Paths doesn\u0026rsquo;t just randomly select instead it is carried out by calculating the packet delivery ratio, residual energy and minimum distance between the source and destination nodes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn encircling stage less optimal routing adjust the parameters by increasing packet delivery ratio and reduced energy consumption in order to prioritize in the next hop that brings the characteristics close to P(t). Let \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\left(t\\right)\\)\u003c/span\u003e\u003c/span\u003e be the best solution found so far, distance in path space between the current path and best known solution and it is provided in Eq.\u0026nbsp;(\u003cspan refid=\"Equ5\" class=\"InternalRef\"\u003e6\u003c/span\u003e)\u003cdiv id=\"Equ5\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ5\" name=\"EquationSource\"\u003e\n$$\\:{D}_{i}\\left(t\\right)=|C*X\\left(t\\right)-\\:{X}_{i}\\left(t\\right)|$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e6\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere C is a constant for randomization factor, for each iteration t\u0026thinsp;+\u0026thinsp;1 and exploration factor A, the update rule in Eq.\u0026nbsp;(\u003cspan refid=\"Equ6\" class=\"InternalRef\"\u003e7\u003c/span\u003e) is defined as\u003cdiv id=\"Equ6\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ6\" name=\"EquationSource\"\u003e\n$$\\:{X}_{i}\\left(t+1\\right)=X\\left(t\\right)-A*{D}_{i}\\left(t\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e7\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere the exploration factor linearly decreases over time with respect to the controls over convergence.\u003c/p\u003e \u003cp\u003eSpiral bubble net provides an exploitative strategy to refine local search around the best optimal route from source to base station to reduce energy depletion, equally divides the load distribution in terms of data integration and provide a reliable communication establishment between source and destination. The ith update of the spiral bubble is provided in Eq.\u0026nbsp;(\u003cspan refid=\"Equ7\" class=\"InternalRef\"\u003e8\u003c/span\u003e)\u003cdiv id=\"Equ7\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ7\" name=\"EquationSource\"\u003e\n$$\\:{X}_{i}\\left(t+1\\right)=X\\left(t\\right)+{D}_{i}\\left(t\\right).{e}^{al}.Cos\\left(2\\varPi\\:l\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e8\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u0026#119886; denotes the logarithmic narrowing constant, while \u0026#119897; refers to a random scalar value within the range [\u0026minus;\u0026thinsp;1,1]. To enable self-motivated swapping between exploration and exploitation phases, Eq.\u0026nbsp;\u003cspan refid=\"Equ8\" class=\"InternalRef\"\u003e9\u003c/span\u003e provide a position routing update mechanism is directed by the following probabilistic rule\u003cdiv id=\"Equ8\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ8\" name=\"EquationSource\"\u003e\n$$\\:{X}_{i}\\left(t+1\\right)=\\left\\{\\begin{array}{c}X\\left(t\\right)-A*{D}_{i}\\left(t\\right)\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:if\\:p\u0026lt;0.5\\\\\\:X\\left(t\\right)+{D}_{i}\\left(t\\right).{e}^{al}.Cos\\left(2\\varPi\\:l\\right)\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:if\\:p\\ge\\:0.5\\end{array}\\right.$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e9\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere p \u0026euro; [0,1]\u003c/p\u003e \u003cp\u003eOnce the optimal solution is produced, the source node communicates the aggregated data to base station. Porpoise algorithm out smart among heuristic intelligent algorithms owing to its frequency tuning capability, enabling automatic changeovers between global and local search. This dynamic control over search correlation, combined simple framework, quick scaling, parallel processing capabilities, ease of use, and great parallel stability, makes it a highly efficient solution.\u003c/p\u003e \u003cp\u003eData integration in a POA-SAE-based wireless sensor network proceeds as follows: First, all network components, including common, routing, and cluster head nodes, are initialized to determine their current states. Next, the monitoring region's clustering structure is established using the sensing node's cluster score. A cluster head node is then randomly selected from each cluster based on its CH Score, and it collects data from its cluster members. Once the WSN clustering stabilizes, the cluster head creates a routing table for shared nodes and communicates member information to the receiving node.\u003c/p\u003e \u003cp\u003eOnce optimal route is established, the WSN's data integration model undergoes training. This process determines the amount of nodes, weightages, and threshold parameters for its middle learning network. To conserve limited energy of shared sensor nodes and extend the network's lifespan, this SAE-based data integration model training is exclusively performed at receiving node. Receiving node first builds the SAE's network structure using the received information. It then trains this network by matching samples from the cluster head sample database with associate node information. After training, the Sink node obtains the data integration model's parameters. Subsequently, receiving node sends these SAE model network parameters to the relevant cluster head. Each cluster head then uses this trained data integration model to fuse data from its member nodes, extracting features and removing redundant or useless information. The merged data is temporarily stored before being sent back to receiving node. Algorithm proposed in this paper, leverages cluster head node feature extraction level and the softmax classification level to handle feature extraction, data integration, and classification. Processed information is then forward to the receiving node.\u003c/p\u003e \u003cp\u003eBased on the Porpoise Optimization Algorithm Stacked Auto encoder (POA-SAE) algorithm, wireless sensor networks follow these multipath routing steps: First, an optimized data integration model is built using stacked auto encoders. To execute multi path routing, this integration model is subsequently incorporated into dynamic heterogeneous wireless sensor network's clustering framework. Enhancing information integration accuracy, lowering data transmission volume, and extending network lifetime are the objectives.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Performance Evaluation","content":"\u003cp\u003eThis study introduces the POA-SAE method, which uses NS3 Simulator for experimental and performance comparison analysis is done through MATLAB to reflect the performance of data integration and multipath routing. assuming that there are enough nodes deployed to ensure a directed node distribution in the monitored area and that a large-scale random node deployment is carried out initially. With 300 sensors, the sensor nodes are set up in a 200x200 m\u003csup\u003e2\u003c/sup\u003e two-dimensional area. There are 50 simulations run in total, with 10 data packets sent from the source to the destination. With an initial node energy of 5J, each packet has a 4 kb capacity. The sensor nodes initialization consumption can reach an average of 0.5 J per node.\u003c/p\u003e \u003cp\u003eThe porpoise optimization mathematical model provides composite WSN witness function, experiment design and formula metrics and statistical analysis method. Porpoise Optimization Algorithm (POA) is a recent bio-inspired method which models alternating exploration and exploitation behaviours. The Performance of POA-SAE is compared with other benchmark algorithms.\u003c/p\u003e \u003cp\u003eThe models for a specific network size were simulated 50 times for each case. The results were collected for the clustering time, total energy consumption, latency, Packet delivery ratio, and the remaining energy in the nodes. Each Sensor node is considered as the porpoise which represent the candidate solution in the problem space. Eq.\u0026nbsp;\u003cspan refid=\"Equ9\" class=\"InternalRef\"\u003e10\u003c/span\u003e denotes search agent of each sensor node.\u003cdiv id=\"Equ9\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ9\" name=\"EquationSource\"\u003e\n$$\\:{X}_{i}^{t}=\\left[{X}_{i,1}^{t},{X}_{i,2}^{t},{X}_{i,3}^{t},\\dots\\:\\dots\\:{X}_{i,D}^{t},\\right]$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e10\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eN denotes number of sensor nodes with value of i varies from i\u0026thinsp;=\u0026thinsp;1,2,3\u0026hellip; N, D dimension of the optimization problem and t represent the iteration index.\u003c/p\u003e \u003cp\u003eThe initial population of the sensor nodes are generated randomly within the spiral bubble of the search limits and is denoted in Eq.\u0026nbsp;\u003cspan refid=\"Equ10\" class=\"InternalRef\"\u003e11\u003c/span\u003e\u003cdiv id=\"Equ10\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ10\" name=\"EquationSource\"\u003e\n$$\\:{x}_{i,d}^{0}=U({L}_{d},{U}_{d})$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e11\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{d},{U}_{d}\\)\u003c/span\u003e\u003c/span\u003e are the lower and upper bounds with dimension d\u003c/p\u003e \u003cp\u003eDefault best solution is initialized as represented in Eq.\u0026nbsp;\u003cspan refid=\"Equ11\" class=\"InternalRef\"\u003e12\u003c/span\u003e\u003cdiv id=\"Equ11\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ11\" name=\"EquationSource\"\u003e\n$$\\:{X}_{B}^{0}=\\text{arg}\\text{min}f\\left({X}_{i}^{0}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e12\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eAfter generating new candidates, the fitness is evaluated to provide a best solution. The \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{i}^{th}\\)\u003c/span\u003e\u003c/span\u003efitness is evaluated in Eq.\u0026nbsp;\u003cspan refid=\"Equ12\" class=\"InternalRef\"\u003e13\u003c/span\u003e.\u003cdiv id=\"Equ12\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ12\" name=\"EquationSource\"\u003e\n$$\\:{F}_{i}^{t+1}=f\\left({X}_{i}^{t+1}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e13\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIf \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{F}_{i}^{t+1}\u0026lt;\\:{F}_{i}^{t}\\)\u003c/span\u003e\u003c/span\u003e then the new position is updated using the Eq.\u0026nbsp;\u003cspan refid=\"Equ13\" class=\"InternalRef\"\u003e14\u003c/span\u003e otherwise the old position is retained, the global best is updated.\u003cdiv id=\"Equ13\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ13\" name=\"EquationSource\"\u003e\n$$\\:{X}_{B}^{t+1}=\\text{arg}\\text{min}{F}_{i}^{t+1}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e14\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e4.1Simulation Results\u003c/h2\u003e \u003cp\u003eConvergence plots the best objective value with respect to number of iterations or rounds. It shows how quickly the optimization algorithm finds the best solution. Eq.\u0026nbsp;\u003cspan refid=\"Equ14\" class=\"InternalRef\"\u003e15\u003c/span\u003e provide the convergence of POA-SAE algorithm.\u003cdiv id=\"Equ14\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ14\" name=\"EquationSource\"\u003e\n$$\\:F=\\:{w}_{1}.{E}_{R}+{w}_{2}.{R}_{i}+{w}_{3}.\\frac{1}{L}+{w}_{4}.\\frac{1}{{C}_{t}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e15\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{w}_{1},\\:{w}_{2},{w}_{3},{w}_{4}\\)\u003c/span\u003e\u003c/span\u003e are weights to ensure the balance\u003c/p\u003e \u003cp\u003eThe convergence behaviour of several algorithms is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. POA-SAE shows a better capacity to identify optimal solutions. POA-SAE exhibits fast convergence; by iterations 20\u0026ndash;25, its fitness curve had risen significantly and stabilized close to the global optimum. Its balanced exploration and exploitation strategies, which enable the algorithm avoid becoming stranded in local minima like a porpoise's encircling and bubble-net tactics, are responsible for this effective convergence. Other algorithms, in contrast, show distinct patterns of convergence. Particle Swarm Optimization (PSO) flattens its fitness curve at a lower value than POA's because it converges faster than Genetic Algorithm (GA) but frequently stalls in local optima. Because GA's crossover and mutation activities are inherently random, it exhibits a slower, more gradual convergence.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe Low-Energy Adaptive Clustering Hierarchy (LEACH) methodology has a flat, constant performance curve with no true optimization as a baseline. While GA and PSO climb at a considerably slower rate, POA-SAE's fitness increases significantly during the first Phase, or iterations 0\u0026ndash;10, as it rapidly improves its cluster-head and route selection. POA-SAE improves its solutions and stabilizes its fitness close to the ideal value during the second phase iterations, which range from 10 to 30. PSO might prematurely plateau during this period, while GA keeps getting better, albeit more slowly.\u003c/p\u003e \u003cp\u003eFor the Final Phase iterations from 30 to 50 POA-SAE's fitness curve remains flat at a high, stable value. GA's fitness continues to improve but ultimately settles at a lower value than POA, while LEACH remains constant throughout the simulation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Residual Energy\u003c/h2\u003e \u003cp\u003eResidual energy represents the remaining battery power of sensor nodes after each communication round i.e. sensing, transmitting, processing and receiving. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e illustrates the residual energy trend versus total number of rounds. Average residual energy of sensor node relating to rounds is calculated using the Eq.\u0026nbsp;\u003cspan refid=\"Equ15\" class=\"InternalRef\"\u003e16\u003c/span\u003e\u003cdiv id=\"Equ15\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ15\" name=\"EquationSource\"\u003e\n$$\\:{E}_{R}^{i}\\left(t\\right)={E}_{0}^{i}-\\sum\\:_{t=1}^{n}\\left({E}_{TX}\\left(p,d\\right)+{E}_{RX}\\left(k\\right)+{E}_{S}+{E}_{p}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e16\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{E}_{0}^{i}\\)\u003c/span\u003e\u003c/span\u003e is the initial node energy i.e. 5J, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{E}_{TX,},\\:{E}_{RX},\\:{E}_{S},\\:\\\u0026amp;\\:{E}_{p}\\)\u003c/span\u003e\u003c/span\u003e are the consumption of transmission, receiving, sensing and processing with respect to packet size p and distance d.\u003c/p\u003e \u003cp\u003eBased on the above provided data in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, POA-SAE protocol exhibits superior energy efficiency compared to PSO, GA, WOA, and LEACH. The residual energy analysis shows that POA-SAE consistently maintains a higher level of energy throughout the simulation. After 50 rounds, POA nodes retain approximately 3.2 J of energy, whereas PSO and GA nodes drop to about 2.9 J and 2.7 J, respectively. This suggests that POA-SAE effectively selects more energy-efficient cluster-heads and routing paths. This energy conservation is attributed to POA-SAE's balanced approach, which incorporates exploration for diverse cluster-head selection and exploitation to focus on optimal clusters. POA-SAE prolongs the stability period of the network and delays the occurrence of dead nodes, thereby extending network lifetime.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eReduced redundancy of sensed information before communicating it to base station, data aggregation lowers transmission latency and energy usage. POA-SAE achieve better cluster head placement and aggregate more efficiently. Graph mentioned in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows the effectiveness of data aggregation under different optimization algorithms. POA - SAE maintains the lowest aggregation ratio 0.4 to 0.5, indicating higher redundancy removal and reduced transmissions compared to PSO of about 0.55, GA 0.6, and LEACH at 0.75. LEACH has less aggregation when compared to our proposed POA-SAE optimization algorithm. Lower ratios mean that fewer redundant packets reach the base station, thus saving energy and extending network lifetime. The result confirms that POA-SAE not only balances energy consumption but also improves data aggregation efficiency.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe clustering time directly influences the stability and responsiveness of a wireless sensor network (WSN). Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e shows the relationship between clustering time and the number of nodes for different optimization algorithms POA-SAE, PSO, GA, and LEACH. The results show that POA- SAE consistently achieves the lowest clustering time compared to PSO, GA, and LEACH. This is due to the effective balance of exploration and exploitation in POA-SAE\u0026rsquo;s search dynamics, which accelerates the convergence towards optimal CH selection. In contrast, LEACH shows the highest clustering time since it relies on randomized CH election, which introduces higher overhead. PSO and GA demonstrate moderate clustering performance but are outperformed by POA-SAE. These results confirm the suitability of POA-SAE for large-scale deployments where fast reconfiguration is essential.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e demonstrates that POA-SAE achieves the most balanced cluster distribution with the lowest variance in cluster sizes. PSO and GA produce reasonably distributed clusters, but slight imbalances remain due to their tendency to converge prematurely. LEACH, on the other hand, exhibits the worst cluster distribution, leading to network hotspots and reduced stability. The balanced distribution achieved by POA-SAE can be attributed to its adaptive exploitation mechanism, which prevents over-concentration of nodes around specific CHs. This ensures fair resource allocation and longer network lifetime.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows how latency changes as the number of rounds increases. The lowest latency is attained via POA-SAE, suggesting effective routing and little congestion. The porpoise-inspired search strategy's optimal path construction is responsible for this improvement. PSO and GA, in contrast, exhibit significant latency reductions but struggle with delayed convergence. Because of its uneven cluster sizes and redundant communication channels, LEACH has the highest latency. POA-SAE's lower latency provides quicker data delivery, which makes it ideal for real-time WSN applications. Considering most of assessed evaluations, POA-SAE shows a notable performance improvement over the baseline algorithms PSO, GA, and LEACH. POA SAE's adaptive exploitation mechanism and spiral movement-based search method, which improves both local and global optimization capability, are responsible for this improvement. The outcomes also demonstrate that POA-SAE is a scalable solution since it can continue to provide consistent performance even as the number of nodes rises. As demonstrated in earlier tests, the enhanced cluster distribution and decreased latency also directly lead to a higher packet delivery ratio and residual energy conservation. The proposed method improves data aggregation and energy efficiency in wireless sensor networks for monitoring regions of different scales in simulations. Therefore, it can be concluded that the strategy proposed in this study effectively improves the overall performance of wireless sensor networks.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eMonitoring several environmental features, including temperature, humidity, pressure, lighting, etc., WSN is made up of numerous dispersed wireless sensor nodes. The paper uses a variety of techniques to enhance energy efficiency, node coverage in WSNs, which automatically increases network\u0026rsquo;s performance and efficiency. The primary research methodology includes data aggregation, energy efficiency analysis, and the application of POA-SAE algorithms to enhance node deployment, which in turn enhances network efficiency and performance. The study's findings demonstrated that POA-SAE significantly impacted optimization. Wireless Sensor Networks (WSNs) are critical for real time monitoring in various domains but face persistent challenges of energy limitation, redundant data transmission, and inefficient routing. To address these, we propose Porpoise Optimization Algorithm (POA) optimized Stacked Autoencoder (SAE)-based multi-hop routing protocol for efficient data integration Network efficiency is significantly enhanced by combining the global search capacity of POA for the best cluster head selection and routing with the synergy of stacked autoencoders for feature extraction and redundancy elimination.\u003c/p\u003e \u003cp\u003eThe proposed POA-SAE methodology is clearly superior to baseline techniques like LEACH, GA, and PSO, as shown by the simulation results. With nodes maintaining roughly 3.2 J after 50 rounds as opposed to 2.9 J in PSO and 2.7 J in GA, POA-SAE continuously maintained higher residual energy, indicating nearly 15\u0026ndash;20% superior energy efficiency. In terms of data aggregation, POA-SAE achieved an average aggregation ratio of 0.4\u0026ndash;0.5, substantially lower than PSO (0.55), GA (0.6), and LEACH (0.75), indicating superior redundancy reduction. Latency was also minimized, with POA-SAE demonstrating up to a 30\u0026ndash;35% reduction compared to baseline methods. Furthermore, clustering time was the lowest across all tested scenarios, confirming its scalability for large-scale WSNs. Overall, POA-SAE delivers improved energy efficiency, scalability, reliability, and faster convergence, making it a robust protocol for critical applications. Future research can extend this framework by incorporating adaptive security mechanisms and hybrid optimization approaches to further strengthen resilience and adaptability in dynamic WSN environments.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cp\u003e6. Funding Declaration\u003c/p\u003e\n\u003cp\u003eThe Authors did not receive funds, grant or other support for the submitted work.\u003c/p\u003e\n\u003cp\u003e7.Conflict of Interest\u003c/p\u003e\n\u003cp\u003eWe declare that no financial or non-financial interest to disclose. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e8. Acknowledgement: None\u003c/p\u003e\n\u003cp\u003e9.Author Contributions:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRamesh D. Provide the conceptualization, analysis and methodology of the research work.\u003c/p\u003e\n\u003cp\u003eJaya T. Provide Suggestion and correction related to methodology and analysis in this research work.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eElhoseny M, Abdelaziz A, Hassanien A (2020) Swarm intelligence\u0026ndash;based energy efficient clustering with multihop routing protocol for sustainable wireless sensor networks. Int J Distrib Sens Netw 16(9):1\u0026ndash;13\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJavaid N, Qureshi TN, Khan AH (2013) EDDEEC: Enhanced developed distributed energy efficient clustering for heterogeneous wireless sensor networks, arXiv preprint arXiv:1303.5274\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQureshi TN, Javaid N, Khan AH (202) Optimized Cluster-Based Dynamic Energy-Aware Routing Protocol for Wireless Sensor Networks in Agriculture Precision. Int. J. Commun. Syst., vol. 33, no. 13, pp. 1\u0026ndash;16\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReddy K (2021) Hybrid optimization algorithm for security aware cluster head selection process to aid hierarchical routing in wireless sensor networks. IET Commun 15(7):821\u0026ndash;832\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMd A, Rahman, Matin MA (2020) Energy-efficient routing in wireless sensor networks: A comprehensive survey. Algorithms, vol. 13, no. 3, pp. 1\u0026ndash;38\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSunithar S, Rajkumar R, Anitha M (2024) An Energy-Efficient and Secure WSN Routing Protocol Using Bayesian Networks and Elitist Genetic Algorithms. J Electr Syst Automat 57(6):610\u0026ndash;619\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGanesh R, Amutha R (2013) Efficient and Secure Routing Protocol for Wireless Sensor Networks through SNR based Dynamic Clustering Mechanisms (ESRPSDC), arXiv preprint arXiv:1306.0312\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen H, Mineno H, Mizuno T (2008) Adaptive data aggregation scheme in clustered wireless sensor networks. Comput Commun 31(15):3579\u0026ndash;3585\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu B, Li J, Li Y (2009) Distributed Data Aggregation Scheduling in Wireless Sensor Networks. In IEEE INFOCOM 2009. IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/infcom.2009.5062140\u003c/span\u003e\u003cspan address=\"10.1109/infcom.2009.5062140\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYe Z, Abouzeid AA, Ai J (2007) Optimal Policies for Distributed Data Aggregation in Wireless Sensor Networks. In \u003cem\u003eIEEE INFOCOM 2007\u0026ndash;26th IEEE International Conference on Computer Communications\u003c/em\u003e (pp. 1676\u0026ndash;1684). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/infcom.2007.196\u003c/span\u003e\u003cspan address=\"10.1109/infcom.2007.196\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLIU M (2005) A Distributed Energy-Efficient Data Gathering and Aggregation Protocol for Wireless Sensor Networks. Journal of Software, vol 16. China Science Publishing \u0026amp; Media Ltd, p 2106. 12 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1360/jos162106\u003c/span\u003e\u003cspan address=\"10.1360/jos162106\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi S, Qi H (2013) Distributed Data Aggregation for Sparse Recovery in Wireless Sensor Networks. In \u003cem\u003e2013 IEEE International Conference on Distributed Computing in Sensor Systems\u003c/em\u003e (pp. 62\u0026ndash;69). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/dcoss.2013.64\u003c/span\u003e\u003cspan address=\"10.1109/dcoss.2013.64\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoubiche DE, Boubiche S, Bilami A, Toral-Cruz H (2016) Toward Adaptive Data Aggregation Protocols in Wireless Sensor Networks. In \u003cem\u003eProceedings of the International Conference on Internet of things and Cloud Computing\u003c/em\u003e (pp. 1\u0026ndash;6). ACM. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2896387.2900325\u003c/span\u003e\u003cspan address=\"10.1145/2896387.2900325\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu L, Yang Z, Li M, Liu D (2013) An Efficient Data Aggregation Protocol Concentrated on Data Integrity in Wireless Sensor Networks. In \u003cem\u003eInternational Journal of Distributed Sensor Networks\u003c/em\u003e (Vol. 9, Issue 5, p. 256852). SAGE Publications. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2013/256852\u003c/span\u003e\u003cspan address=\"10.1155/2013/256852\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArdakani SP (2017) Data Aggregation Routing Protocols in Wireless Sensor Networks: A Taxonomy. In \u003cem\u003eInternational journal of Computer Networks \u0026amp; Communications\u003c/em\u003e (Vol. 9, Issue 2, pp. 89\u0026ndash;107). Academy and Industry Research Collaboration Center (AIRCC). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5121/ijcnc.2017.9207\u003c/span\u003e\u003cspan address=\"10.5121/ijcnc.2017.9207\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJi P, Li Y, Jiang J, Wang T (2012) A Clustering Protocol for Data Aggregation in Wireless Sensor Network. In \u003cem\u003e2012 International Conference on Control Engineering and Communication Technology\u003c/em\u003e (pp. 649\u0026ndash;652). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/iccect.2012.260\u003c/span\u003e\u003cspan address=\"10.1109/iccect.2012.260\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChatterjee S, Pais AR, Saurabh S (2016) Distributed Data Aggregation in Wireless Sensor Network - with Peer Verification. In Proceedings of the 13th International Joint Conference on e-Business and Telecommunications (pp. 440\u0026ndash;445). SCITEPRESS - Science and Technology Publications. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5220/0005991604400445\u003c/span\u003e\u003cspan address=\"10.5220/0005991604400445\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBagaa M, Lasla N, Ouadjaout A, Challal Y (2007) SEDAN: Secure and Efficient protocol for Data Aggregation in wireless sensor Networks. In \u003cem\u003e32nd IEEE Conference on Local Computer Networks (LCN 2007)\u003c/em\u003e (pp. 1053\u0026ndash;1060). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/lcn.2007.81\u003c/span\u003e\u003cspan address=\"10.1109/lcn.2007.81\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl-Yasiri A, Sunley A (2007) Data aggregation in wireless sensor networks using the SOAP protocol. In \u003cem\u003eJournal of Physics: Conference Series\u003c/em\u003e (Vol. 76, p. 012039). IOP Publishing. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1088/1742-6596/76/1/012039\u003c/span\u003e\u003cspan address=\"10.1088/1742-6596/76/1/012039\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBo W, Han-ying H, Wen F (2008) An Improved LEACH Protocol for Data Gathering and Aggregation in Wireless Sensor Networks. In \u003cem\u003e2008 International Conference on Computer and Electrical Engineering\u003c/em\u003e (pp. 398\u0026ndash;401). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/iccee.2008.59\u003c/span\u003e\u003cspan address=\"10.1109/iccee.2008.59\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYANG J, ZHANG ZHANGD-Y, Y.-Y., WANG Y (2010) Cluster-Based Data Aggregation and Transmission Protocol for Wireless Sensor Networks. Journal of Software, vol 21. China Science Publishing \u0026amp; Media Ltd, pp 1127\u0026ndash;1137. 5 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3724/sp.j.1001.2010.03534\u003c/span\u003e\u003cspan address=\"10.3724/sp.j.1001.2010.03534\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSudha C, Suresh D, Nagesh A (2021) An Enhanced Machine learning data aggregation model for Efficient data processing in wireless sensor networks. In \u003cem\u003e2021 6th International Conference on Communication and Electronics Systems (ICCES)\u003c/em\u003e (pp. 1748\u0026ndash;1753). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/icces51350.2021.9489124\u003c/span\u003e\u003cspan address=\"10.1109/icces51350.2021.9489124\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eC S, D, S., A N (2021) Accurate Data Aggregation Created by Neural Network and Data Classification Processed Through Machine Learning in Wireless Sensor Networks. Res Square Platf LLC. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21203/rs.3.rs-895195/v1\u003c/span\u003e\u003cspan address=\"10.21203/rs.3.rs-895195/v1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePellenz ME, Lachowski R, Jamhour E, Brante G, Moritz GL, Souza RD (2021) In-Network Data Aggregation for Information-Centric WSNs using Unsupervised Machine Learning Techniques. In 2021 IEEE Symposium on Computers and Communications (ISCC) (pp. 1\u0026ndash;7). IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/iscc53001.2021.9631416\u003c/span\u003e\u003cspan address=\"10.1109/iscc53001.2021.9631416\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen JL, Summers JE (2016) Deep neural networks for learning classification features and generative models from synthetic aperture sonar big data. In Proceedings of Meetings on Acoustics. Acoustical Society of America. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1121/2.0000458\u003c/span\u003e\u003cspan address=\"10.1121/2.0000458\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIlichev VY, Chukhraev IV (2021) Data processing using deep learning of the generative-adversarial neural network (GAN). In Neurocomputers. Publishing house Radiotekhnika \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18127/j19998554-202105-04\u003c/span\u003e\u003cspan address=\"10.18127/j19998554-202105-04\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu C, Xu G, Ding Y, Zhao J (2019) Explore Deep Neural Network and Reinforcement Learning to Large-scale Tasks Processing in Big Data. International Journal of Pattern Recognition and Artificial Intelligence, vol 33. World Scientific Pub Co Pte Ltd, p 1951010. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1142/s0218001419510108\u003c/span\u003e\u003cspan address=\"10.1142/s0218001419510108\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu T (2023) Data Fusion Algorithm For Wireless Sensor Networks Based On Cluster Routing And Deep Learning. In Journal of Physics: Conference Series (Vol. 2504, Issue 1, p. 012043). IOP Publishing. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1088/1742-6596/2504/1/012043\u003c/span\u003e\u003cspan address=\"10.1088/1742-6596/2504/1/012043\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"peer-to-peer-networking-and-applications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ppna","sideBox":"Learn more about [Peer-to-Peer Networking and Applications](http://link.springer.com/journal/12083)","snPcode":"12083","submissionUrl":"https://submission.nature.com/new-submission/12083/3","title":"Peer-to-Peer Networking and Applications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Scalability, Energy Efficient, Reliability, Network life time, Data aggregation, Redundant data","lastPublishedDoi":"10.21203/rs.3.rs-7799400/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7799400/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRise of Wireless Sensor Networks (WSNs) for applications ranging from environmental monitoring to healthcare, defence, and disaster management is an essential technology in the recent trends. Resource constraints such as energy resources, redundant data transmission, routing inefficiencies, and security vulnerabilities prime to deprived performance. To overcome these restrictions, a Bioinspired algorithm based on Porpoise Optimization Algorithm (POA) optimized deep learning-based multi-hop routing protocol designed for efficient data integration in WSNs. The proposed framework integrates the feature extraction capability of Stacked Auto encoders (SAEs) with the global and local search efficiency of POA to optimize cluster head selection, multipath routing, and data aggregation. In this approach, nodes are clustered into clusters based on remaining energy, node density, and distance, with cluster heads selected using a POA enhanced latent space mechanism. The Porpoise Optimization Algorithm - Stacked Auto encoders (POA-SAE) model compressed the high dimensional sensor data into compact latent representations, thereby minimizing communication overhead, redundancy of data and improving data accuracy. The integration of POA and SAE ensures balanced energy consumption across nodes by dynamically identifying the optimal routing paths through encircling and spiral search strategies inspired by porpoise foraging behaviour. Simulations are carried out in NS3 Whereas performance evaluation is carried out in MATLAB environments to demonstrate that the proposed POA-SAE protocol significantly outperforms conventional approaches such as LEACH, PSO, and GA. Performance metrics including residual energy, clustering time, packet delivery ratio, latency, and data aggregation efficiency expose substantial improvements in scalability, stability, and network lifetime. Specifically, POA-SAE achieves faster convergence, higher residual energy retention, and superior cluster distribution, while effectively reducing transmission delays and redundant packets.\u003c/p\u003e","manuscriptTitle":"Porpoise Optimized Deep Learning Based Multi-Hop Routing Protocol for Efficient Data Integration in WSN","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-29 09:28:06","doi":"10.21203/rs.3.rs-7799400/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-14T09:46:53+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-08T03:03:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"176050880586595990244253271179231534124","date":"2026-04-26T03:38:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"180450920292760733476142059089079867915","date":"2026-04-23T17:37:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"135117311823950068702781321132003090136","date":"2026-04-23T07:19:32+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-21T03:36:46+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-11T17:12:40+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-09T07:27:47+00:00","index":"","fulltext":""},{"type":"submitted","content":"Peer-to-Peer Networking and Applications","date":"2025-10-07T12:05:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"peer-to-peer-networking-and-applications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ppna","sideBox":"Learn more about [Peer-to-Peer Networking and Applications](http://link.springer.com/journal/12083)","snPcode":"12083","submissionUrl":"https://submission.nature.com/new-submission/12083/3","title":"Peer-to-Peer Networking and Applications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"7530ee87-c50b-4531-875c-4203b2398ab3","owner":[],"postedDate":"April 29th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-14T09:46:53+00:00","index":19,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-08T03:03:21+00:00","index":18,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-29T09:28:07+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-29 09:28:06","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7799400","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7799400","identity":"rs-7799400","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}