Enhanced Battery Degradation and RUL Prediction Using Bidirectional LSTM Networks

Enhanced Battery Degradation and RUL Prediction Using Bidirectional LSTM Networks | 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 Enhanced Battery Degradation and RUL Prediction Using Bidirectional LSTM Networks Aman This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6212719/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Accurate prediction of battery degradation and remaining useful life (RUL) is critical for optimizing the performance and lifespan of battery-powered systems in electric vehicles and renewable energy storage applications. This paper introduces a novel machine learning approach utilizing Bidirectional Long Short-Term Memory (Bi-LSTM) networks to predict battery degradation and estimate RUL based on key parameters including voltage, current, temperature, and cycle count. Unlike conventional LSTM models that process data in a unidirectional manner, our Bi-LSTM architecture captures both past and future dependencies in battery behavior, significantly improving prediction accuracy. Through comprehensive evaluation on real-world battery datasets, we demonstrate that Bi-LSTM outperforms traditional LSTM systems by reducing root mean square error (RMSE) for state of health (SOH) prediction from 4.5–3.1% and improving R² values from 0.87 to 0.92. For RUL prediction, our model achieves an RMSE of 120 cycles compared to 150 cycles for standard LSTM. These improvements enable more reliable real-time battery health monitoring and proactive management strategies. The integration of Bi-LSTM into battery management systems (BMS) offers enhanced computational efficiency and superior convergence speed, making it particularly suitable for applications requiring precise battery management such as electric vehicles and grid-scale energy storage systems. Renewable Resources Electronic Materials and Devices Battery Degradation Prediction Remaining Useful Life Estimation Deep Learning in Energy Storage Battery Cycle Life Prediction Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Batteries have become indispensable components in modern energy systems, serving as critical power sources for electric vehicles (EVs), consumer electronics, renewable energy storage, and industrial applications [ 1 ]. Among various battery technologies, lithium-ion batteries (LIBs) are particularly prominent due to their high energy density, long cycle life, and low self-discharge rate [ 2 ]. However, the performance of these batteries degrades over time and with usage, leading to capacity loss and decreased efficiency. This degradation process is influenced by multiple factors including charge-discharge cycles, temperature variations, and operating conditions, creating complex, non-linear patterns that are challenging to model [ 3 ]. Accurate prediction of battery degradation and remaining useful life (RUL) is essential for ensuring the reliability, safety, and cost-effectiveness of battery-powered systems [ 4 ]. Precise RUL estimation enables proactive maintenance, optimized charging strategies, and timely replacement decisions, ultimately extending battery lifespan and improving system performance. Traditional approaches to battery health monitoring often struggle to capture the intricate relationships between various operational parameters and degradation patterns [ 3 ], [ 5 ]. Machine learning techniques have emerged as powerful tools for modeling complex systems and predicting non-linear behaviors. Among these, Long Short-Term Memory (LSTM) networks have shown particular promise for time-series prediction tasks due to their ability to capture long-term dependencies in sequential data [ 6 ]. However, conventional LSTM models process data in a unidirectional manner, limiting their capacity to fully capture both past and future patterns in battery degradation trends [ 7 ]. This research proposes a novel approach utilizing Bidirectional LSTM (Bi-LSTM) networks to predict battery degradation and estimate RUL. By processing data in both forward and backward directions, Bi-LSTM can integrate information from both past and future time steps, providing a more comprehensive understanding of battery behavior. This architecture is particularly well-suited for battery health prediction tasks where degradation patterns may exhibit complex temporal relationships. The key contributions of this work include: Development of a Bi-LSTM-based prediction model that outperforms conventional LSTM systems in capturing bidirectional dependencies in battery degradation data Comprehensive evaluation on real-world battery datasets demonstrating superior prediction accuracy for both state of health (SOH) and RUL Integration of the Bi-LSTM model into battery management systems (BMS) for real-time health monitoring and proactive management Detailed analysis of the model's performance across different battery chemistries and operating conditions The remainder of this paper is organized as follows: Section 2 provides background on battery degradation mechanisms and existing approaches for SOH and RUL estimation. Section 3 details the methodology of the proposed Bi-LSTM model, including data preprocessing, feature engineering, and model architecture. Section 4 presents the experimental results and comparative analysis with conventional methods. Finally, Section 5 offers conclusions and directions for future research. Battery Degradation and State of Health Estimation LIBs are widely used in electric vehicles (EVs), renewable energy systems, and portable electronics, exhibit gradual performance degradation over time and with usage [ 2 ]. This degradation manifests as capacity loss, increased internal resistance, and reduced power capability, ultimately affecting the overall performance and lifespan of battery-powered systems. SOH is a critical metric for assessing battery condition and is defined as the ratio of the battery's current capacity to its nominal capacity under ideal conditions [ 8 ], [ 9 ]. Mathematically, it can be expressed in Eq. ( 1 ). $$\:SOH=\frac{{C}_{current}}{{C}_{nominal}}\times\:100\%$$ 1 where \(\:{C}_{current}\) is the battery's current capacity and \(\:{C}_{nominal}\) is its nominal capacity when new. A battery is typically considered to have reached its end of life when its SOH falls below 70–80%, depending on application requirements [ 8 ]. Accurate real-time SOH estimation is essential for implementing effective battery management strategies, optimizing charging/discharging patterns, predicting RUL, ensuring safety and reliability of battery systems, and reducing overall ownership costs through optimized usage. Estimating SOH accurately presents several challenges. Battery degradation results from multiple interrelated electrochemical processes that vary with operating conditions, temperature, and usage patterns [ 10 ]. The relationship between operational parameters (voltage, current, temperature) and SOH is highly non-linear and difficult to model with simple equations. Battery performance data can be noisy and influenced by measurement errors, environmental factors, and individual cell variations. Furthermore, direct measurement of SOH through complete charge-discharge cycles is time-consuming and impractical for real-time applications. SOH estimation methods can be categorized into three main approaches. Direct measurement methods involve directly measuring battery parameters that correlate with SOH, such as capacity measurement through complete charge-discharge cycles, internal resistance measurement, and differential voltage analysis. While providing direct insights, these methods are often impractical for real-time applications due to time requirements and potential disruption of normal operation [ 11 ]. Model-based approaches use electrochemical or equivalent circuit models to simulate battery behavior and estimate SOH [ 12 ], [ 13 ]. These methods require detailed knowledge of battery internal chemistry, can be computationally intensive, may struggle with parameter identification, and often need calibration for different battery types and conditions. Data-driven approaches leverage historical battery data to train predictive models without requiring detailed knowledge of internal battery chemistry [ 14 ]. These methods can handle complex non-linear relationships, enable real-time estimation once trained, are flexible to different battery types and conditions, but require large amounts of quality training data. While each approach has merits, they also have significant limitations. Many methods rely on features extracted from complete charging profiles, which are not always available in real-world applications [ 15 ]. Traditional LSTM models process data sequentially in one direction, limiting their ability to capture complete temporal patterns [ 16 ]. Some advanced models require significant computational resources, making them unsuitable for embedded systems in battery management applications [ 6 ]. Models trained on specific battery datasets often perform poorly when applied to batteries with different chemistries or operating conditions [ 10 ]. These limitations highlight the need for more advanced, flexible, and computationally efficient methods for SOH estimation that can handle the complex, multi-directional dependencies in battery degradation data. RUL Estimation Approaches Accurate prediction of a battery's RUL is crucial for optimizing battery management systems and extending battery lifespan in applications ranging from electric vehicles to renewable energy storage. Various approaches have been developed to estimate RUL, each with distinct advantages and limitations. Statistical approaches leverage historical degradation data to establish probabilistic models for RUL prediction. These methods often employ probability distributions and stochastic processes to model battery failure mechanisms. For example, a probabilistic model using a Weibull distribution has demonstrated promising results in predicting battery failure times by characterizing the degradation process through statistical parameters [ 17 ]. Bayesian updating methods, combined with particle filtering techniques, have also shown improved forecast accuracy by incorporating uncertainty measurements into the prediction process [ 18 ]. While these statistical approaches provide a quantitative basis for RUL estimation, they typically require large amounts of historical data and may struggle with capturing the complex, non-linear degradation patterns common in real-world battery applications. Physics-based models utilize the fundamental electrochemical and thermal principles governing battery operation to predict degradation. These models incorporate detailed battery chemistry and physics, including electrode degradation, electrolyte decomposition, and thermal effects. For instance, electrochemical models based on porous electrode theory can simulate lithium-ion battery degradation by tracking the evolution of key electrochemical parameters over time [ 12 ]. Thermal-electrochemical coupled models further enhance prediction accuracy by considering how temperature variations influence degradation processes [ 12 ]. While physics-based approaches offer physically meaningful insights and can extrapolate beyond observed data, they require comprehensive knowledge of battery internal chemistry and can be computationally intensive, limiting their applicability in real-time battery management systems. Machine learning techniques have gained popularity for RUL prediction due to their ability to model complex non-linear relationships without requiring explicit physical models. Traditional machine learning algorithms such as Support Vector Regression have been successfully applied to battery RUL prediction, demonstrating robustness to noisy data [ 19 ]. Random Forest models have achieved high prediction accuracy by classifying battery health states based on extracted features from battery operation data [ 20 ]. Artificial neural networks have also been employed, using voltage and current features to estimate RUL with improved generalization capabilities [ 6 ]. These methods typically require careful feature engineering to select relevant health indicators from battery operation data, which can be challenging due to the high dimensionality and noise in battery datasets. Deep learning approaches, particularly recurrent neural networks and their variants such as LSTM networks, have shown significant promise for RUL prediction. These architectures can automatically learn hierarchical representations of battery degradation patterns from sequential data. The LSTM architecture, introduced by Hochreiter and Schmidhuber, established a foundation for time-series prediction tasks by addressing the vanishing gradient problem through specialized memory cells [ 6 ]. Jafari et al. developed an LSTM-based model for predictive maintenance, demonstrating superior performance over traditional machine learning approaches [ 21 ]. Sohn et al. applied LSTM networks to battery degradation prediction, achieving high accuracy by leveraging long-term dependencies in sequential battery data [ 22 ]. More recently, Zhao et al. integrated Bi-LSTM networks with attention mechanisms, enhancing model interpretability and prediction stability [ 23 ]. While deep learning approaches offer powerful modeling capabilities, they typically require substantial training data and computational resources, and may suffer from overfitting if not properly regularized [ 24 ], [ 25 ]. Despite advances in RUL prediction techniques, several challenges remain. Battery degradation patterns exhibit significant variability due to diverse operating conditions, manufacturing inconsistencies, and environmental factors [ 22 ]. Identifying optimal health indicators (HIs) remains difficult due to noise and redundancy in battery datasets [ 26 ]. Computational efficiency is another concern, as many advanced models require extensive training and hyperparameter tuning, increasing computational costs [ 10 ]. Generalization across different battery chemistries and operating conditions remains limited, as models trained on specific datasets often perform poorly when applied to batteries with different characteristics [ 10 ]. Addressing these challenges requires developing more robust, efficient, and adaptable RUL prediction methods that can handle the complexity and variability of real-world battery degradation data. Existing System The current system for battery degradation prediction primarily utilizes LSTM networks, which have been widely adopted for their effectiveness in handling time-series data and capturing long-term dependencies [ 14 ], [ 15 ]. This approach represents the conventional methodology against which our proposed Bi-LSTM system will be compared and improved upon. System Architecture and Workflow The LSTM-based system follows a structured workflow beginning with data collection from reliable sources such as the Prognostics Center of Excellence battery dataset. The raw battery data undergoes comprehensive preprocessing, which includes cleaning to remove noise and artifacts, normalization to standardize the data scale, and handling of missing values through interpolation or other appropriate strategies [ 12 ]. Feature Selection and Engineering Health indicators (HIs) are carefully selected based on empirical ratio tests and domain knowledge. These indicators typically include voltage, current, temperature, and cycle count, which are extracted from the battery operation data. The dataset is then segmented into fixed-length time windows to create input sequences for the LSTM model [ 19 ]. LSTM Model Architecture The LSTM model architecture consists of several layers: Input Layer : Accepts a sequence of battery parameters including voltage, current, temperature, and other relevant operational data. LSTM Layer : Captures both short-term fluctuations and long-term degradation trends through its specialized memory cells and gating mechanisms [ 19 ]. Dense (Fully Connected) Layer : Maps the LSTM outputs to a continuous RUL prediction. Output Layer : Provides the final estimated RUL value, typically expressed in terms of remaining charge cycles. Model Training and Optimization The LSTM model is trained using the Mean Squared Error (MSE) loss function, which quantifies the difference between predicted and actual RUL values. Optimizers such as Adam or Stochastic Gradient Descent are employed to adjust the model parameters and minimize the loss function [ 6 ]. Hyperparameter tuning is performed to optimize the model performance, with parameters including the number of LSTM layers, the number of units in each layer, and dropout rates being systematically adjusted and evaluated. RUL Prediction and Evaluation Once trained, the LSTM model is tested on unseen battery data to assess its generalization capabilities. Performance metrics such as Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R² Score are calculated to evaluate the prediction accuracy. If the error metrics exceed acceptable thresholds, the hyperparameters are readjusted to refine the model performance. System Limitations Despite its effectiveness in many applications, the existing LSTM-based system exhibits several limitations: Sensitivity to Parameter Variations: LSTMs can exhibit high variance in predictions due to minor fluctuations in battery parameters. Computational Demands: Training deep LSTM networks requires substantial computational resources and time, particularly when processing large battery datasets [ 6 ]. Feature Selection Challenges: Identifying the most relevant and effective health indicators (HIs) remains a significant challenge due to the high dimensionality of battery data and the presence of noise and redundancy [ 12 ]. Unidirectional Processing: Traditional LSTM models process data sequentially in one direction, limiting their ability to fully capture the complete temporal context of battery degradation patterns [ 16 ]. These limitations motivate the development of our proposed Bi-LSTM system, which addresses these issues through bidirectional processing, improved feature selection methods, and enhanced computational efficiency while maintaining or improving prediction accuracy. Proposed System Our proposed system leverages a Bi-LSTM architecture to predict battery degradation and estimate RUL by analyzing both past and future temporal contexts within battery operation data. This approach addresses the limitations of conventional LSTM models by capturing bidirectional dependencies in battery degradation patterns, thereby significantly enhancing prediction accuracy and reliability. 1. Data Collection and Preprocessing The foundation of our system is built upon comprehensive datasets that include critical battery parameters such as voltage, current, temperature, and charge-discharge cycles. The preprocessing stage involves rigorous data cleaning to remove noise and artifacts, normalization to standardize data scales, and sophisticated handling of missing values through advanced interpolation techniques [ 8 ], [ 9 ]. This stage ensures that the input data maintains the highest possible quality for model training and validation. 2. Feature Engineering and Health Indicators Selection We extract meaningful health indicators (HIs) from the raw battery data, focusing on parameters such as capacity degradation rate, internal resistance, and charge-discharge efficiency. These HIs are selected through a combination of domain knowledge and empirical analysis. To further enhance model performance, we employ a Binary Firefly Optimization Algorithm (BFA) to identify the most relevant features for predicting SOH and RUL [ 11 ]. This optimization process effectively reduces noise and redundancy in the dataset while improving computational efficiency. 3. Bidirectional LSTM Model Architecture Our Bi-LSTM model architecture is designed to capture both past and future dependencies in battery degradation patterns. As illustrated in Fig. 1 , the model consists of: Input Layer: Accepts time-series battery data including voltage, current, and temperature measurements [ 16 ]. Bi-LSTM Layers: Processes data in both forward and backward directions, enabling the model to analyze long-term dependencies from multiple temporal perspectives [ 17 ]. Fully Connected Layer: Extracts high-level features from the bidirectional outputs. Output Layer: Provides the final prediction of battery RUL [ 12 ]. The model utilizes Mean Squared Error (MSE) as the loss function to optimize prediction accuracy [ 13 ], with the Adam Optimizer employed for its efficient learning capabilities [ 19 ]. 4. Model Evaluation and Execution Comparison We conduct comprehensive evaluations comparing our Bi-LSTM model against conventional LSTM systems. The results demonstrate that our proposed architecture achieves superior accuracy in capturing both past and future patterns in battery degradation data. Specifically, our model exhibits faster convergence during training and lower prediction errors in both SOH and RUL estimation tasks [ 20 ]. Statistical validation methods including Mann-Whitney U test and Dynamic Time Warping (DTW) are applied to confirm data alignment and model robustness. Cross-validation techniques ensure the generalization capabilities of our model across different battery datasets and operating conditions [ 21 ]. 5. Deployment and Integration into Battery Management System The trained Bi-LSTM model is integrated into a comprehensive BMS framework. This integration enables real-time RUL prediction and degradation monitoring, providing early warnings for potential battery failure with unprecedented accuracy. The system continuously analyzes battery operation parameters such as voltage, current, and temperature to optimize charging and discharging cycles [ 17 ]. This proactive approach not only extends battery lifespan but also enhances overall system safety and reliability. As depicted in Fig. 2 , the proposed system follows a structured workflow that begins with data collection and preprocessing, followed by feature engineering and model training. The trained model is then deployed within the BMS for real-time monitoring and prediction, enabling proactive battery management strategies. Our proposed Bi-LSTM system offers several significant advantages over conventional approaches: Bidirectional Context Capture: By processing data in both directions, the model gains a more complete understanding of battery degradation patterns, leading to more accurate predictions [ 22 ]. Enhanced Feature Utilization: The Binary Firefly Optimization Algorithm ensures that only the most relevant features are used, reducing computational overhead while improving model performance. Real-Time Monitoring Capability: Integration with BMS allows for continuous assessment and immediate response to changing battery conditions [ 10 ]. Improved Prediction Stability: The architecture reduces error propagation and captures subtle variations in battery performance over time, resulting in more reliable RUL estimates. Continuing forward: Bi-LSTM prisoners the Bidleen position, and runs for more accurate roll forecasts [ 22 ]. Skilled highlight fixation, The BFA guarantees because it was most important that it is used, a toll is taken to reduce calculation. Early joining, Bi-LSTM reduces the preparation of time while maintaining longer prime performance. Real time check, Integration into BMS allows continuous examination and presentation help [ 10 ]. The proposed Bi-LSTM system represents a substantial advancement in battery degradation prediction and RUL estimation. By addressing the limitations of conventional LSTM models through bidirectional processing and optimized feature selection, our system achieves superior prediction accuracy while maintaining computational efficiency. This approach enables more effective battery management strategies, ultimately extending battery lifespan and enhancing the performance of battery-powered systems in critical applications such as electric vehicles and renewable energy storage. Results and Discussion This section presents the analysis of battery degradation using our proposed Bi-LSTM model. We evaluated the model's performance using comprehensive datasets that capture various battery degradation patterns under different operating conditions. Figure 3 compares the actual SOH values with those predicted by our Bi-LSTM model across different battery cycles. The blue line represents the actual SOH values, while the orange line indicates the Bi-LSTM predictions. The close alignment between the predicted and actual SOH values demonstrates the model's high accuracy in capturing the temporal differences in battery health indicators. Figure 4 illustrates how the SOH estimate improves after applying proper changes in the output by the trained model. The figure shows the MSE between the preparations, indicating that the SOH predictions are currently more adjusted with real values, demonstrating that the work on the connected predictions and error model has made a difference in progress. Table 1 presents the predicted degradation pattern of a lithium-ion battery across various charging-discharging cycles. The results show a consistent and logical degradation trend, with the battery capacity decreasing as the number of cycles increases. The Bi-LSTM model accurately predicts the SOH values, with minor deviations that can be further minimized through hyperparameter tuning. Table 1. Predicted degradation pattern of a lithium-ion battery across charging-discharging cycles Cycle Capacity (Ah) SOH (%) Degradation (%) 50 1.8 100.0 0.0 100 1.5 83.3 16.7 200 1.2 66.7 33.3 300 1.0 55.6 44.4 400 0.8 44.4 55.6 Figure 5 compares the actual RUL values with those predicted by our Bi-LSTM model. The blue line represents the actual RUL, while the orange line shows the Bi-LSTM predictions. The model demonstrates strong performance in capturing the long-term degradation trends, with predictions closely following the actual RUL values. Table 2 provides an analysis of battery capacity degradation over charge-discharge cycles and the corresponding predicted RUL in cycles. The results indicate that our model can reliably estimate how many cycles remain before the battery reaches its end of life. Table 2. Analysis of battery capacity degradation and predicted RUL across charge-discharge cycles Cycle Capacity (Ah) Predicted RUL (Cycles) 50 1.8 ~450 100 1.5 ~400 200 1.2 ~300 300 1.0 ~200 400 0.8 ~100 Figure 6 is shown the output of the RUL by using the Bi-LSTM which compares the remaining life expected of the battery. The performance provides a stable expectation and confirms that Bi-LSTM can guess regularly with low errors. Table 3 presents a comparison of SOH predictions for a battery across different charge-discharge cycles using both LSTM and Bi-LSTM models, compared against the actual SOH values. The Bi-LSTM model shows significantly closer alignment with actual values across all cycles. Table 3. Comparison of SOH predictions using LSTM and BI-LSTM models across different charge-discharge cycles Cycle Actual SOH LSTM SOH Bi-LSTM SOH 50 1.00 1.05 1.02 100 0.92 0.95 0.92 200 0.82 0.85 0.81 300 0.72 0.75 0.71 400 0.62 0.65 0.61 Table 4 presents the degradation in SOH of a battery over different cycles as predicted by both LSTM and Bi-LSTM models, compared with the actual degradation values. The Bi-LSTM predictions consistently align better with the actual degradation rates. Table 4. Degradation in SOH predicted by LSTM and BI-LSTM models compared with actual values. LSTM Degradation Bi-LSTM Degradation Actual Degradation -0.05 -0.02 0.00 0.05 0.08 0.08 0.15 0.19 0.18 0.25 0.29 0.28 0.35 0.39 0.38 Table 5 compares the performance of the LSTM and Bi-LSTM models using RMSE and R² (R-Squared) values for both SOH and RUL predictions. The Bi-LSTM model demonstrates superior performance across all metrics. Discussion Table 5. Performance comparison of LSTM and BI-LSTM models for SOH and RUL prediction using RMSE and R² metrics Model RMSE (SOH) R² (SOH) RMSE (SOH) R² (SOH) LSTM 4.5% 0.87 150 cycles 0.85 Bi-LSTM 3.1% 0.92 120 cycles 0.91 The comparative analysis between the LSTM and Bi-LSTM models highlights the superior performance of our proposed Bi-LSTM architecture in predicting battery degradation and RUL. The results show that Bi-LSTM achieves lower error rates and higher accuracy compared to conventional LSTM. Specifically, for SOH prediction, Bi-LSTM reduces the RMSE from 4.5% (LSTM) to 3.1% and improves the R² value from 0.87 to 0.92. For RUL prediction, Bi-LSTM achieves an RMSE of 120 cycles compared to 150 cycles for LSTM, while improving the R² value from 0.85 to 0.91. These improvements can be attributed to the Bi-LSTM's ability to capture both past and future dependencies in battery degradation patterns. This bidirectional processing allows the model to better understand the complex temporal relationships in battery data, leading to more accurate predictions. The enhanced accuracy and lower error margins make Bi-LSTM a promising approach for real-time applications in BMS. This improved prediction capability enables better maintenance strategies, optimized charging cycles, and extended battery lifespan. Conclusion The proposed Bi-LSTM model demonstrates superior performance in predicting battery degradation and estimating RUL compared to conventional LSTM approaches. Our comprehensive evaluation using real-world battery datasets shows that Bi-LSTM significantly improves prediction accuracy by capturing bidirectional dependencies in battery degradation patterns. For SOH prediction, our model reduces the RMSE from 4.5% (LSTM) to 3.1% and improves the R² value from 0.87 to 0.92. In RUL prediction, Bi-LSTM achieves an RMSE of 120 cycles compared to 150 cycles for standard LSTM, while improving the R² value from 0.85 to 0.91. These enhancements are particularly significant for applications requiring precise battery management, such as electric vehicles and renewable energy storage systems. The integration of Bi-LSTM into BMS enables real-time monitoring and more accurate health assessment, allowing for optimized charging strategies, proactive maintenance scheduling, and extended battery lifespan. This results in reduced operational costs and improved safety for battery-powered systems. Future work will focus on expanding the model to handle multiple battery chemistries simultaneously, incorporating additional sensor data such as impedance spectroscopy, and developing adaptive versions that can update in real-time as new data becomes available. We also plan to explore transfer learning approaches to apply knowledge from well-characterized batteries to new or less-characterized ones, further broadening the applicability of this technology. References B. Dunn, H. Kamath, and J.-M. Tarascon, “Electrical Energy Storage for the Grid: A Battery of Choices,” Science , vol. 334, no. 6058, pp. 928–935, Nov. 2011, doi: 10.1126/science.1212741. L. Kouchachvili, W. Yaïci, and E. Entchev, “Hybrid battery/supercapacitor energy storage system for the electric vehicles,” J. Power Sources , vol. 374, pp. 237–248, Jan. 2018, doi: 10.1016/j.jpowsour.2017.11.040. B. Ospina Agudelo, W. Zamboni, and E. Monmasson, “Application domain extension of incremental capacity-based battery SoH indicators,” Energy , vol. 234, p. 121224, Nov. 2021, doi: 10.1016/j.energy.2021.121224. J. Tian, R. Xiong, W. Shen, and J. Lu, “Data-driven battery degradation prediction: Forecasting voltage-capacity curves using one-cycle data,” EcoMat , vol. 4, no. 5, p. e12213, 2022, doi: 10.1002/eom2.12213. Aman and R. S. Chhillar, “The Upsurge of Deep Learning for Disease Prediction in Healthcare,” in Innovations in Data Analytics , in Advances in Intelligent Systems and Computing. Singapore: Springer Nature, 2023, pp. 511–518. doi: 10.1007/978-981-99-0550-8_40. N. Costa, D. Anseán, M. Dubarry, and L. Sánchez, “ICFormer: A Deep Learning model for informed lithium-ion battery diagnosis and early knee detection,” J. Power Sources , vol. 592, p. 233910, Feb. 2024, doi: 10.1016/j.jpowsour.2023.233910. L. Mao, J. Xu, J. Chen, J. Zhao, Y. Wu, and F. Yao, “A LSTM-STW and GS-LM Fusion Method for Lithium-Ion Battery RUL Prediction Based on EEMD,” Energies , vol. 13, no. 9, 2020, doi: 10.3390/en13092380. P. Shi, R. Hu, Y. He, X. Chen, H. Huang, and Z. Liu, “A novel LFSSA-LSTM algorithm for the remaining useful life prediction of lithium-ion batteries in whole life cycle,” Int. J. Electrochem. Sci. , vol. 19, no. 9, p. 100719, Sep. 2024, doi: 10.1016/j.ijoes.2024.100719. C. Lin, X. Tuo, L. Wu, G. Zhang, and X. Zeng, “Accurate Capacity Prediction and Evaluation with Advanced SSA-CNN-BiLSTM Framework for Lithium-Ion Batteries,” Batteries , vol. 10, no. 3, Art. no. 3, Mar. 2024, doi: 10.3390/batteries10030071. J. Xia et al. , “Historical data-independent remaining useful life prediction method based on dual-input deep learning neural network,” J. Energy Storage , vol. 72, p. 108427, Nov. 2023, doi: 10.1016/j.est.2023.108427. P. Shi, R. Hu, Y. He, X. Chen, H. Huang, and Z. Liu, “A novel LFSSA-LSTM algorithm for the remaining useful life prediction of lithium-ion batteries in whole life cycle,” Int. J. Electrochem. Sci. , vol. 19, no. 9, p. 100719, Sep. 2024, doi: 10.1016/j.ijoes.2024.100719. C. Han and H. Yuan, “Capacity prediction of Lithium-ion batteries based on adaptive sliding window pooling extreme learning machine algorithm,” Phys. Scr. , vol. 99, no. 1, p. 016003, Jan. 2024, doi: 10.1088/1402-4896/ad1704. J. Dou et al. , “Extreme learning machine model for state-of-charge estimation of lithium-ion battery using salp swarm algorithm,” J. Energy Storage , vol. 52, p. 104996, Aug. 2022, doi: 10.1016/j.est.2022.104996. Y. Ji, Z. Chen, Y. Shen, K. Yang, Y. Wang, and J. Cui, “An RUL prediction approach for lithium-ion battery based on SADE-MESN,” Appl Soft Comput , vol. 104, no. C, Jun. 2021, doi: 10.1016/j.asoc.2021.107195. X. Bian, Z. Wei, J. He, F. Yan, and L. Liu, “A Novel Model-Based Voltage Construction Method for Robust State-of-Health Estimation of Lithium-Ion Batteries,” IEEE Trans. Ind. Electron. , vol. 68, no. 12, pp. 12173–12184, Dec. 2021, doi: 10.1109/TIE.2020.3044779. X. Zhang, H. Li, W. Zhang, A. M. Lopes, X. Wu, and L. Chen, “Improved Parameter Identification for Lithium-Ion Batteries Based on Complex-Order Beetle Swarm Optimization Algorithm,” Micromachines , vol. 14, no. 2, p. 413, Feb. 2023, doi: 10.3390/mi14020413. T. Yuan, N. Li, H. Sun, and S. Tan, “A Joint Estimation Method of SOC and SOH for Lithium-ion Battery Considering Cyber-Attacks Based on GA-BP,” Comput. Mater. Contin. , vol. 80, no. 3, pp. 4497–4512, 2024, doi: 10.32604/cmc.2024.056061. S. Zhao, C. Zhang, and Y. Wang, “Lithium-ion battery capacity and remaining useful life prediction using board learning system and long short-term memory neural network,” J. Energy Storage , vol. 52, p. 104901, Aug. 2022, doi: 10.1016/j.est.2022.104901. V. Safavi, A. Mohammadi Vaniar, N. Bazmohammadi, J. C. Vasquez, and J. M. Guerrero, “Battery Remaining Useful Life Prediction Using Machine Learning Models: A Comparative Study,” Information , vol. 15, no. 3, 2024, doi: 10.3390/info15030124. H. Zhao, Z. Chen, X. Shu, J. Shen, Z. Lei, and Y. Zhang, “State of health estimation for lithium-ion batteries based on hybrid attention and deep learning,” Reliab. Eng. Syst. Saf. , vol. 232, p. 109066, Apr. 2023, doi: 10.1016/j.ress.2022.109066. S. Jafari and Y.-C. Byun, “A CNN-GRU Approach to the Accurate Prediction of Batteries’ Remaining Useful Life from Charging Profiles,” Computers , vol. 12, no. 11, 2023, doi: 10.3390/computers12110219. X. Qiao, Z. Wang, X. Su, G. Shen, E. Hou, and Y. Cai, “Hybrid approach for online capacity estimation of lithium-lon batteries: Integrating model-driven and data-driven methods,” J. Energy Storage , vol. 113, p. 115582, Mar. 2025, doi: 10.1016/j.est.2025.115582. V. Yamaçli, “State-of-health estimation and classification of series-connected batteries by using deep learning based hybrid decision approach,” Heliyon , vol. 10, no. 20, p. e39121, Oct. 2024, doi: 10.1016/j.heliyon.2024.e39121. Aman and R. S. Chhillar, “Disease Prediction using Deep Learning Algorithms in Healthcare Sector,” in Disease Prediction using Deep Learning Algorithms in Healthcare Sector , 108-115: Technoarete Publishing, 2022. doi: 10.36647/MLAIDA/2022.12.B1.Ch008. A. Darolia, R. S. Chhillar, M. Alhussein, S. Dalal, K. Aurangzeb, and U. K. Lilhore, “Enhanced cardiovascular disease prediction through self-improved Aquila optimized feature selection in quantum neural network & LSTM model,” Front. Med. , vol. 11, Jun. 2024, doi: 10.3389/fmed.2024.1414637. S. Jafari, J.-H. Yang, and Y.-C. Byun, “Optimized XGBoost modeling for accurate battery capacity degradation prediction,” Results Eng. , vol. 24, p. 102786, Dec. 2024, doi: 10.1016/j.rineng.2024.102786. Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted 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-6212719","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":427865232,"identity":"0fd36714-cc83-455b-b308-f52a739b3295","order_by":0,"name":"Aman","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAkUlEQVRIiWNgGAWjYFACxsYHDwyANDMJWpoNEkjUwsAmkUCSs+Tbm9sqEgoOyzGw8x4gTovBmYNtNxIMDhszMPMRaZmBRCJIS1piAzOPAZEOm/+wrQCopZ54LQw3GNsYEgxsEhiI1mJwJrFZAqjFsI14h7Uff/jhwx8JeX7+M8Q6DAbYSFQ/CkbBKBgFowAfAAAO9CPkuuXnzgAAAABJRU5ErkJggg==","orcid":"","institution":"M.D. University","correspondingAuthor":true,"prefix":"","firstName":"","middleName":"","lastName":"Aman","suffix":""}],"badges":[],"createdAt":"2025-03-12 13:54:57","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":true,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":true},"doi":"10.21203/rs.3.rs-6212719/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6212719/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":78501534,"identity":"b14b785a-48fb-4705-b8b8-292034c7d5c9","added_by":"auto","created_at":"2025-03-14 06:44:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":141420,"visible":true,"origin":"","legend":"\u003cp\u003eBlock diagram of the BI-LSTM layer architecture for battery degradation and RUL prediction\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/73f9c8ac624db65a46d80787.png"},{"id":78501530,"identity":"d5472fe7-499e-4b39-9b8b-c7cb02a4e242","added_by":"auto","created_at":"2025-03-14 06:44:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":176769,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of the proposed BI-LSTM-based system for battery degradation and remaining useful life prediction\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/394a9f16cff8f773dbd2f7c0.png"},{"id":78501531,"identity":"16880b86-cee0-404d-b1fa-ab739ebe34fc","added_by":"auto","created_at":"2025-03-14 06:44:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":149952,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of actual SOH values with BI-LSTM predictions across battery cycles\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/11615a6d3c38be7008a81b6d.png"},{"id":78501814,"identity":"aaff7878-716d-45fb-9ca1-20e8c4b084c7","added_by":"auto","created_at":"2025-03-14 06:52:07","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":56611,"visible":true,"origin":"","legend":"\u003cp\u003eImprovement in SOH estimation accuracy after model optimization and training\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/276f668501b3609852c5198e.png"},{"id":78501813,"identity":"d20ef498-7fb0-45a3-af44-62c0a308a5a1","added_by":"auto","created_at":"2025-03-14 06:52:07","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":135570,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of actual RUL values with BI-LSTM predictions across battery cycles\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/2646b056c52337e9e26f295a.png"},{"id":78501536,"identity":"389d3eef-d716-4ec9-8de8-c2bb7c526d7b","added_by":"auto","created_at":"2025-03-14 06:44:07","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":76663,"visible":true,"origin":"","legend":"\u003cp\u003eStable RUL prediction output using the BI-LSTM model for battery life estimation\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/7c51fbf7f2e3de3ffdfb817c.png"},{"id":78503043,"identity":"a0df9b5b-cfcb-4e82-8d01-763775d77083","added_by":"auto","created_at":"2025-03-14 07:16:08","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1313140,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6212719/v1/03c86f4e-34cb-47f7-ab51-5d542e0eb871.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eEnhanced Battery Degradation and RUL Prediction Using Bidirectional LSTM Networks\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eBatteries have become indispensable components in modern energy systems, serving as critical power sources for electric vehicles (EVs), consumer electronics, renewable energy storage, and industrial applications [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Among various battery technologies, lithium-ion batteries (LIBs) are particularly prominent due to their high energy density, long cycle life, and low self-discharge rate [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, the performance of these batteries degrades over time and with usage, leading to capacity loss and decreased efficiency. This degradation process is influenced by multiple factors including charge-discharge cycles, temperature variations, and operating conditions, creating complex, non-linear patterns that are challenging to model [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAccurate prediction of battery degradation and remaining useful life (RUL) is essential for ensuring the reliability, safety, and cost-effectiveness of battery-powered systems [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Precise RUL estimation enables proactive maintenance, optimized charging strategies, and timely replacement decisions, ultimately extending battery lifespan and improving system performance. Traditional approaches to battery health monitoring often struggle to capture the intricate relationships between various operational parameters and degradation patterns [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMachine learning techniques have emerged as powerful tools for modeling complex systems and predicting non-linear behaviors. Among these, Long Short-Term Memory (LSTM) networks have shown particular promise for time-series prediction tasks due to their ability to capture long-term dependencies in sequential data [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. However, conventional LSTM models process data in a unidirectional manner, limiting their capacity to fully capture both past and future patterns in battery degradation trends [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis research proposes a novel approach utilizing Bidirectional LSTM (Bi-LSTM) networks to predict battery degradation and estimate RUL. By processing data in both forward and backward directions, Bi-LSTM can integrate information from both past and future time steps, providing a more comprehensive understanding of battery behavior. This architecture is particularly well-suited for battery health prediction tasks where degradation patterns may exhibit complex temporal relationships.\u003c/p\u003e \u003cp\u003eThe key contributions of this work include:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eDevelopment of a Bi-LSTM-based prediction model that outperforms conventional LSTM systems in capturing bidirectional dependencies in battery degradation data\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eComprehensive evaluation on real-world battery datasets demonstrating superior prediction accuracy for both state of health (SOH) and RUL\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIntegration of the Bi-LSTM model into battery management systems (BMS) for real-time health monitoring and proactive management\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eDetailed analysis of the model's performance across different battery chemistries and operating conditions\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe remainder of this paper is organized as follows: Section 2 provides background on battery degradation mechanisms and existing approaches for SOH and RUL estimation. Section 3 details the methodology of the proposed Bi-LSTM model, including data preprocessing, feature engineering, and model architecture. Section 4 presents the experimental results and comparative analysis with conventional methods. Finally, Section 5 offers conclusions and directions for future research.\u003c/p\u003e"},{"header":"Battery Degradation and State of Health Estimation","content":"\u003cp\u003eLIBs are widely used in electric vehicles (EVs), renewable energy systems, and portable electronics, exhibit gradual performance degradation over time and with usage [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This degradation manifests as capacity loss, increased internal resistance, and reduced power capability, ultimately affecting the overall performance and lifespan of battery-powered systems.\u003c/p\u003e \u003cp\u003eSOH is a critical metric for assessing battery condition and is defined as the ratio of the battery's current capacity to its nominal capacity under ideal conditions [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Mathematically, it can be expressed in Eq.\u0026nbsp;(\u003cspan refid=\"Equ1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:SOH=\\frac{{C}_{current}}{{C}_{nominal}}\\times\\:100\\%$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{C}_{current}\\)\u003c/span\u003e\u003c/span\u003e is the battery's current capacity and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{C}_{nominal}\\)\u003c/span\u003e\u003c/span\u003e is its nominal capacity when new.\u003c/p\u003e \u003cp\u003eA battery is typically considered to have reached its end of life when its SOH falls below 70–80%, depending on application requirements [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Accurate real-time SOH estimation is essential for implementing effective battery management strategies, optimizing charging/discharging patterns, predicting RUL, ensuring safety and reliability of battery systems, and reducing overall ownership costs through optimized usage.\u003c/p\u003e \u003cp\u003eEstimating SOH accurately presents several challenges. Battery degradation results from multiple interrelated electrochemical processes that vary with operating conditions, temperature, and usage patterns [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. The relationship between operational parameters (voltage, current, temperature) and SOH is highly non-linear and difficult to model with simple equations. Battery performance data can be noisy and influenced by measurement errors, environmental factors, and individual cell variations. Furthermore, direct measurement of SOH through complete charge-discharge cycles is time-consuming and impractical for real-time applications.\u003c/p\u003e \u003cp\u003eSOH estimation methods can be categorized into three main approaches. Direct measurement methods involve directly measuring battery parameters that correlate with SOH, such as capacity measurement through complete charge-discharge cycles, internal resistance measurement, and differential voltage analysis. While providing direct insights, these methods are often impractical for real-time applications due to time requirements and potential disruption of normal operation [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eModel-based approaches use electrochemical or equivalent circuit models to simulate battery behavior and estimate SOH [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. These methods require detailed knowledge of battery internal chemistry, can be computationally intensive, may struggle with parameter identification, and often need calibration for different battery types and conditions.\u003c/p\u003e \u003cp\u003eData-driven approaches leverage historical battery data to train predictive models without requiring detailed knowledge of internal battery chemistry [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. These methods can handle complex non-linear relationships, enable real-time estimation once trained, are flexible to different battery types and conditions, but require large amounts of quality training data.\u003c/p\u003e \u003cp\u003eWhile each approach has merits, they also have significant limitations. Many methods rely on features extracted from complete charging profiles, which are not always available in real-world applications [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Traditional LSTM models process data sequentially in one direction, limiting their ability to capture complete temporal patterns [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Some advanced models require significant computational resources, making them unsuitable for embedded systems in battery management applications [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Models trained on specific battery datasets often perform poorly when applied to batteries with different chemistries or operating conditions [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. These limitations highlight the need for more advanced, flexible, and computationally efficient methods for SOH estimation that can handle the complex, multi-directional dependencies in battery degradation data.\u003c/p\u003e "},{"header":"RUL Estimation Approaches","content":"\u003cp\u003eAccurate prediction of a battery's RUL is crucial for optimizing battery management systems and extending battery lifespan in applications ranging from electric vehicles to renewable energy storage. Various approaches have been developed to estimate RUL, each with distinct advantages and limitations.\u003c/p\u003e\u003cp\u003eStatistical approaches leverage historical degradation data to establish probabilistic models for RUL prediction. These methods often employ probability distributions and stochastic processes to model battery failure mechanisms. For example, a probabilistic model using a Weibull distribution has demonstrated promising results in predicting battery failure times by characterizing the degradation process through statistical parameters [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Bayesian updating methods, combined with particle filtering techniques, have also shown improved forecast accuracy by incorporating uncertainty measurements into the prediction process [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. While these statistical approaches provide a quantitative basis for RUL estimation, they typically require large amounts of historical data and may struggle with capturing the complex, non-linear degradation patterns common in real-world battery applications.\u003c/p\u003e\u003cp\u003ePhysics-based models utilize the fundamental electrochemical and thermal principles governing battery operation to predict degradation. These models incorporate detailed battery chemistry and physics, including electrode degradation, electrolyte decomposition, and thermal effects. For instance, electrochemical models based on porous electrode theory can simulate lithium-ion battery degradation by tracking the evolution of key electrochemical parameters over time [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Thermal-electrochemical coupled models further enhance prediction accuracy by considering how temperature variations influence degradation processes [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. While physics-based approaches offer physically meaningful insights and can extrapolate beyond observed data, they require comprehensive knowledge of battery internal chemistry and can be computationally intensive, limiting their applicability in real-time battery management systems.\u003c/p\u003e\u003cp\u003eMachine learning techniques have gained popularity for RUL prediction due to their ability to model complex non-linear relationships without requiring explicit physical models. Traditional machine learning algorithms such as Support Vector Regression have been successfully applied to battery RUL prediction, demonstrating robustness to noisy data [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Random Forest models have achieved high prediction accuracy by classifying battery health states based on extracted features from battery operation data [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Artificial neural networks have also been employed, using voltage and current features to estimate RUL with improved generalization capabilities [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. These methods typically require careful feature engineering to select relevant health indicators from battery operation data, which can be challenging due to the high dimensionality and noise in battery datasets.\u003c/p\u003e\u003cp\u003eDeep learning approaches, particularly recurrent neural networks and their variants such as LSTM networks, have shown significant promise for RUL prediction. These architectures can automatically learn hierarchical representations of battery degradation patterns from sequential data. The LSTM architecture, introduced by Hochreiter and Schmidhuber, established a foundation for time-series prediction tasks by addressing the vanishing gradient problem through specialized memory cells [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Jafari et al. developed an LSTM-based model for predictive maintenance, demonstrating superior performance over traditional machine learning approaches [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Sohn et al. applied LSTM networks to battery degradation prediction, achieving high accuracy by leveraging long-term dependencies in sequential battery data [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. More recently, Zhao et al. integrated Bi-LSTM networks with attention mechanisms, enhancing model interpretability and prediction stability [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. While deep learning approaches offer powerful modeling capabilities, they typically require substantial training data and computational resources, and may suffer from overfitting if not properly regularized [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDespite advances in RUL prediction techniques, several challenges remain. Battery degradation patterns exhibit significant variability due to diverse operating conditions, manufacturing inconsistencies, and environmental factors [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Identifying optimal health indicators (HIs) remains difficult due to noise and redundancy in battery datasets [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Computational efficiency is another concern, as many advanced models require extensive training and hyperparameter tuning, increasing computational costs [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Generalization across different battery chemistries and operating conditions remains limited, as models trained on specific datasets often perform poorly when applied to batteries with different characteristics [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Addressing these challenges requires developing more robust, efficient, and adaptable RUL prediction methods that can handle the complexity and variability of real-world battery degradation data.\u003c/p\u003e"},{"header":"Existing System","content":"\u003cp\u003eThe current system for battery degradation prediction primarily utilizes LSTM networks, which have been widely adopted for their effectiveness in handling time-series data and capturing long-term dependencies [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. This approach represents the conventional methodology against which our proposed Bi-LSTM system will be compared and improved upon.\u003c/p\u003e\u003ch3\u003eSystem Architecture and Workflow\u003c/h3\u003e\u003cp\u003eThe LSTM-based system follows a structured workflow beginning with data collection from reliable sources such as the Prognostics Center of Excellence battery dataset. The raw battery data undergoes comprehensive preprocessing, which includes cleaning to remove noise and artifacts, normalization to standardize the data scale, and handling of missing values through interpolation or other appropriate strategies [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\u003ch3\u003eFeature Selection and Engineering\u003c/h3\u003e\u003cp\u003eHealth indicators (HIs) are carefully selected based on empirical ratio tests and domain knowledge. These indicators typically include voltage, current, temperature, and cycle count, which are extracted from the battery operation data. The dataset is then segmented into fixed-length time windows to create input sequences for the LSTM model [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e\u003ch3\u003eLSTM Model Architecture\u003c/h3\u003e\u003cp\u003eThe LSTM model architecture consists of several layers:\u003c/p\u003e\u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eInput Layer\u003c/b\u003e: Accepts a sequence of battery parameters including voltage, current, temperature, and other relevant operational data.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eLSTM Layer\u003c/b\u003e: Captures both short-term fluctuations and long-term degradation trends through its specialized memory cells and gating mechanisms [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eDense (Fully Connected) Layer\u003c/b\u003e: Maps the LSTM outputs to a continuous RUL prediction.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eOutput Layer\u003c/b\u003e: Provides the final estimated RUL value, typically expressed in terms of remaining charge cycles.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e\u003ch2\u003eModel Training and Optimization\u003c/h2\u003e\u003cp\u003eThe LSTM model is trained using the Mean Squared Error (MSE) loss function, which quantifies the difference between predicted and actual RUL values. Optimizers such as Adam or Stochastic Gradient Descent are employed to adjust the model parameters and minimize the loss function [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Hyperparameter tuning is performed to optimize the model performance, with parameters including the number of LSTM layers, the number of units in each layer, and dropout rates being systematically adjusted and evaluated.\u003c/p\u003e\u003ch3\u003eRUL Prediction and Evaluation\u003c/h3\u003e\u003cp\u003eOnce trained, the LSTM model is tested on unseen battery data to assess its generalization capabilities. Performance metrics such as Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R² Score are calculated to evaluate the prediction accuracy. If the error metrics exceed acceptable thresholds, the hyperparameters are readjusted to refine the model performance.\u003c/p\u003e\u003ch3\u003eSystem Limitations\u003c/h3\u003e\u003cp\u003eDespite its effectiveness in many applications, the existing LSTM-based system exhibits several limitations:\u003c/p\u003e\u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eSensitivity to Parameter Variations: LSTMs can exhibit high variance in predictions due to minor fluctuations in battery parameters.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eComputational Demands: Training deep LSTM networks requires substantial computational resources and time, particularly when processing large battery datasets [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eFeature Selection Challenges: Identifying the most relevant and effective health indicators (HIs) remains a significant challenge due to the high dimensionality of battery data and the presence of noise and redundancy [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eUnidirectional Processing: Traditional LSTM models process data sequentially in one direction, limiting their ability to fully capture the complete temporal context of battery degradation patterns [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e\u003cp\u003eThese limitations motivate the development of our proposed Bi-LSTM system, which addresses these issues through bidirectional processing, improved feature selection methods, and enhanced computational efficiency while maintaining or improving prediction accuracy.\u003c/p\u003e"},{"header":"Proposed System","content":"\u003cp\u003eOur proposed system leverages a Bi-LSTM architecture to predict battery degradation and estimate RUL by analyzing both past and future temporal contexts within battery operation data. This approach addresses the limitations of conventional LSTM models by capturing bidirectional dependencies in battery degradation patterns, thereby significantly enhancing prediction accuracy and reliability.\u003c/p\u003e\u003ch2\u003e1. Data Collection and Preprocessing\u003c/h2\u003e\u003cp\u003eThe foundation of our system is built upon comprehensive datasets that include critical battery parameters such as voltage, current, temperature, and charge-discharge cycles. The preprocessing stage involves rigorous data cleaning to remove noise and artifacts, normalization to standardize data scales, and sophisticated handling of missing values through advanced interpolation techniques [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. This stage ensures that the input data maintains the highest possible quality for model training and validation.\u003c/p\u003e\u003ch2\u003e2. Feature Engineering and Health Indicators Selection\u003c/h2\u003e\u003cp\u003eWe extract meaningful health indicators (HIs) from the raw battery data, focusing on parameters such as capacity degradation rate, internal resistance, and charge-discharge efficiency. These HIs are selected through a combination of domain knowledge and empirical analysis. To further enhance model performance, we employ a Binary Firefly Optimization Algorithm (BFA) to identify the most relevant features for predicting SOH and RUL [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. This optimization process effectively reduces noise and redundancy in the dataset while improving computational efficiency.\u003c/p\u003e\u003ch2\u003e3. Bidirectional LSTM Model Architecture\u003c/h2\u003e\u003cp\u003eOur Bi-LSTM model architecture is designed to capture both past and future dependencies in battery degradation patterns. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the model consists of:\u003c/p\u003e\u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eInput Layer: Accepts time-series battery data including voltage, current, and temperature measurements [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eBi-LSTM Layers: Processes data in both forward and backward directions, enabling the model to analyze long-term dependencies from multiple temporal perspectives [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eFully Connected Layer: Extracts high-level features from the bidirectional outputs.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eOutput Layer: Provides the final prediction of battery RUL [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e\u003cp\u003eThe model utilizes Mean Squared Error (MSE) as the loss function to optimize prediction accuracy [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], with the Adam Optimizer employed for its efficient learning capabilities [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e\u003ch2\u003e4. Model Evaluation and Execution Comparison\u003c/h2\u003e\u003cp\u003eWe conduct comprehensive evaluations comparing our Bi-LSTM model against conventional LSTM systems. The results demonstrate that our proposed architecture achieves superior accuracy in capturing both past and future patterns in battery degradation data. Specifically, our model exhibits faster convergence during training and lower prediction errors in both SOH and RUL estimation tasks [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Statistical validation methods including Mann-Whitney U test and Dynamic Time Warping (DTW) are applied to confirm data alignment and model robustness. Cross-validation techniques ensure the generalization capabilities of our model across different battery datasets and operating conditions [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e\u003ch2\u003e5. Deployment and Integration into Battery Management System\u003c/h2\u003e\u003cp\u003eThe trained Bi-LSTM model is integrated into a comprehensive BMS framework. This integration enables real-time RUL prediction and degradation monitoring, providing early warnings for potential battery failure with unprecedented accuracy. The system continuously analyzes battery operation parameters such as voltage, current, and temperature to optimize charging and discharging cycles [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. This proactive approach not only extends battery lifespan but also enhances overall system safety and reliability.\u003c/p\u003e\u003cp\u003eAs depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the proposed system follows a structured workflow that begins with data collection and preprocessing, followed by feature engineering and model training. The trained model is then deployed within the BMS for real-time monitoring and prediction, enabling proactive battery management strategies.\u003c/p\u003e\u003cp\u003eOur proposed Bi-LSTM system offers several significant advantages over conventional approaches:\u003c/p\u003e\u003cul\u003e \u003cli\u003e \u003cp\u003eBidirectional Context Capture: By processing data in both directions, the model gains a more complete understanding of battery degradation patterns, leading to more accurate predictions [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEnhanced Feature Utilization: The Binary Firefly Optimization Algorithm ensures that only the most relevant features are used, reducing computational overhead while improving model performance.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eReal-Time Monitoring Capability: Integration with BMS allows for continuous assessment and immediate response to changing battery conditions [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eImproved Prediction Stability: The architecture reduces error propagation and captures subtle variations in battery performance over time, resulting in more reliable RUL estimates.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eContinuing forward: Bi-LSTM prisoners the Bidleen position, and runs for more accurate roll forecasts [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Skilled highlight fixation, The BFA guarantees because it was most important that it is used, a toll is taken to reduce calculation. Early joining, Bi-LSTM reduces the preparation of time while maintaining longer prime performance. Real time check, Integration into BMS allows continuous examination and presentation help [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e\u003cp\u003eThe proposed Bi-LSTM system represents a substantial advancement in battery degradation prediction and RUL estimation. By addressing the limitations of conventional LSTM models through bidirectional processing and optimized feature selection, our system achieves superior prediction accuracy while maintaining computational efficiency. This approach enables more effective battery management strategies, ultimately extending battery lifespan and enhancing the performance of battery-powered systems in critical applications such as electric vehicles and renewable energy storage.\u003c/p\u003e"},{"header":"Results and Discussion","content":"\u003cp\u003eThis section presents the analysis of battery degradation using our proposed Bi-LSTM model. We evaluated the model\u0026apos;s performance using comprehensive datasets that capture various battery degradation patterns under different operating conditions.\u003c/p\u003e\n\u003cp\u003eFigure 3 compares the actual SOH values with those predicted by our Bi-LSTM model across different battery cycles. The blue line represents the actual SOH values, while the orange line indicates the Bi-LSTM predictions. The close alignment between the predicted and actual SOH values demonstrates the model\u0026apos;s high accuracy in capturing the temporal differences in battery health indicators.\u003c/p\u003e\n\u003cp\u003eFigure 4 illustrates how the SOH estimate improves after applying proper changes in the output by the trained model. The figure shows the MSE between the preparations, indicating that the SOH predictions are currently more adjusted with real values, demonstrating that the work on the connected predictions and error model has made a difference in progress.\u003c/p\u003e\n\u003cp\u003eTable 1 presents the predicted degradation pattern of a lithium-ion battery across various charging-discharging cycles. The results show a consistent and logical degradation trend, with the battery capacity decreasing as the number of cycles increases. The Bi-LSTM model accurately predicts the SOH values, with minor deviations that can be further minimized through hyperparameter tuning.\u003c/p\u003e\n\u003cp\u003eTable 1. Predicted degradation pattern of a lithium-ion battery across charging-discharging cycles\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCycle\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCapacity (Ah)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSOH (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eDegradation (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e100.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e83.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e16.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e200\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e66.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e33.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e55.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e44.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e400\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e44.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e55.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eFigure 5 compares the actual RUL values with those predicted by our Bi-LSTM model. The blue line represents the actual RUL, while the orange line shows the Bi-LSTM predictions. The model demonstrates strong performance in capturing the long-term degradation trends, with predictions closely following the actual RUL values.\u003c/p\u003e\n\u003cp\u003eTable 2 provides an analysis of battery capacity degradation over charge-discharge cycles and the corresponding predicted RUL in cycles. The results indicate that our model can reliably estimate how many cycles remain before the battery reaches its end of life.\u003c/p\u003e\n\u003cp\u003eTable 2. Analysis of battery capacity degradation and predicted RUL across charge-discharge cycles\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003eCycle\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003eCapacity (Ah)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003ePredicted RUL (Cycles)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e50\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e1.8\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e~450\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e100\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e1.5\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e~400\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e200\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e1.2\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e~300\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e300\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e1.0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e~200\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e400\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e0.8\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003e~100\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eFigure 6 is shown the output of the RUL by using the Bi-LSTM which compares the remaining life expected of the battery. The performance provides a stable expectation and confirms that Bi-LSTM can guess regularly with low errors.\u003c/p\u003e\n\u003cp\u003eTable 3 presents a comparison of SOH predictions for a battery across different charge-discharge cycles using both LSTM and Bi-LSTM models, compared against the actual SOH values. The Bi-LSTM model shows significantly closer alignment with actual values across all cycles.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 3. Comparison of SOH predictions using LSTM and BI-LSTM models across different charge-discharge cycles\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCycle\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eActual SOH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLSTM SOH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBi-LSTM SOH\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e200\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e400\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.61\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eTable 4 presents the degradation in SOH of a battery over different cycles as predicted by both LSTM and Bi-LSTM models, compared with the actual degradation values. The Bi-LSTM predictions consistently align better with the actual degradation rates.\u003c/p\u003e\n\u003cp\u003eTable 4. Degradation in SOH predicted by LSTM and BI-LSTM models compared with actual values.\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLSTM Degradation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBi-LSTM Degradation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eActual Degradation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eTable 5 compares the performance of the LSTM and Bi-LSTM models using RMSE and R\u0026sup2; (R-Squared) values for both SOH and RUL predictions. The Bi-LSTM model demonstrates superior performance across all metrics.\u003c/p\u003e\n\u003cp\u003eDiscussion\u003c/p\u003e\n\u003cp\u003eTable 5. Performance comparison of LSTM and BI-LSTM models for SOH and RUL prediction using RMSE and R\u0026sup2; metrics\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRMSE (SOH)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;R\u0026sup2; (SOH)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRMSE (SOH)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eR\u0026sup2; (SOH)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e150 cycles\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBi-LSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e120 cycles\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe comparative analysis between the LSTM and Bi-LSTM models highlights the superior performance of our proposed Bi-LSTM architecture in predicting battery degradation and RUL. The results show that Bi-LSTM achieves lower error rates and higher accuracy compared to conventional LSTM. Specifically, for SOH prediction, Bi-LSTM reduces the RMSE from 4.5% (LSTM) to 3.1% and improves the R\u0026sup2; value from 0.87 to 0.92. For RUL prediction, Bi-LSTM achieves an RMSE of 120 cycles compared to 150 cycles for LSTM, while improving the R\u0026sup2; value from 0.85 to 0.91.\u003c/p\u003e\n\u003cp\u003eThese improvements can be attributed to the Bi-LSTM\u0026apos;s ability to capture both past and future dependencies in battery degradation patterns. This bidirectional processing allows the model to better understand the complex temporal relationships in battery data, leading to more accurate predictions.\u003c/p\u003e\n\u003cp\u003eThe enhanced accuracy and lower error margins make Bi-LSTM a promising approach for real-time applications in BMS. This improved prediction capability enables better maintenance strategies, optimized charging cycles, and extended battery lifespan.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe proposed Bi-LSTM model demonstrates superior performance in predicting battery degradation and estimating RUL compared to conventional LSTM approaches. Our comprehensive evaluation using real-world battery datasets shows that Bi-LSTM significantly improves prediction accuracy by capturing bidirectional dependencies in battery degradation patterns. For SOH prediction, our model reduces the RMSE from 4.5% (LSTM) to 3.1% and improves the R\u0026sup2; value from 0.87 to 0.92. In RUL prediction, Bi-LSTM achieves an RMSE of 120 cycles compared to 150 cycles for standard LSTM, while improving the R\u0026sup2; value from 0.85 to 0.91. These enhancements are particularly significant for applications requiring precise battery management, such as electric vehicles and renewable energy storage systems. The integration of Bi-LSTM into BMS enables real-time monitoring and more accurate health assessment, allowing for optimized charging strategies, proactive maintenance scheduling, and extended battery lifespan. This results in reduced operational costs and improved safety for battery-powered systems. Future work will focus on expanding the model to handle multiple battery chemistries simultaneously, incorporating additional sensor data such as impedance spectroscopy, and developing adaptive versions that can update in real-time as new data becomes available. We also plan to explore transfer learning approaches to apply knowledge from well-characterized batteries to new or less-characterized ones, further broadening the applicability of this technology.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eB. Dunn, H. Kamath, and J.-M. Tarascon, \u0026ldquo;Electrical Energy Storage for the Grid: A Battery of Choices,\u0026rdquo; \u003cem\u003eScience\u003c/em\u003e, vol. 334, no. 6058, pp. 928\u0026ndash;935, Nov. 2011, doi: 10.1126/science.1212741.\u003c/li\u003e\n\u003cli\u003eL. Kouchachvili, W. Ya\u0026iuml;ci, and E. Entchev, \u0026ldquo;Hybrid battery/supercapacitor energy storage system for the electric vehicles,\u0026rdquo; \u003cem\u003eJ. Power Sources\u003c/em\u003e, vol. 374, pp. 237\u0026ndash;248, Jan. 2018, doi: 10.1016/j.jpowsour.2017.11.040.\u003c/li\u003e\n\u003cli\u003eB. Ospina Agudelo, W. Zamboni, and E. Monmasson, \u0026ldquo;Application domain extension of incremental capacity-based battery SoH indicators,\u0026rdquo; \u003cem\u003eEnergy\u003c/em\u003e, vol. 234, p. 121224, Nov. 2021, doi: 10.1016/j.energy.2021.121224.\u003c/li\u003e\n\u003cli\u003eJ. Tian, R. Xiong, W. Shen, and J. Lu, \u0026ldquo;Data-driven battery degradation prediction: Forecasting voltage-capacity curves using one-cycle data,\u0026rdquo; \u003cem\u003eEcoMat\u003c/em\u003e, vol. 4, no. 5, p. e12213, 2022, doi: 10.1002/eom2.12213.\u003c/li\u003e\n\u003cli\u003eAman and R. S. Chhillar, \u0026ldquo;The Upsurge of Deep Learning for Disease Prediction in Healthcare,\u0026rdquo; in \u003cem\u003eInnovations in Data Analytics\u003c/em\u003e, in Advances in Intelligent Systems and Computing. Singapore: Springer Nature, 2023, pp. 511\u0026ndash;518. doi: 10.1007/978-981-99-0550-8_40.\u003c/li\u003e\n\u003cli\u003eN. Costa, D. Anse\u0026aacute;n, M. Dubarry, and L. S\u0026aacute;nchez, \u0026ldquo;ICFormer: A Deep Learning model for informed lithium-ion battery diagnosis and early knee detection,\u0026rdquo; \u003cem\u003eJ. Power Sources\u003c/em\u003e, vol. 592, p. 233910, Feb. 2024, doi: 10.1016/j.jpowsour.2023.233910.\u003c/li\u003e\n\u003cli\u003eL. Mao, J. Xu, J. Chen, J. Zhao, Y. Wu, and F. Yao, \u0026ldquo;A LSTM-STW and GS-LM Fusion Method for Lithium-Ion Battery RUL Prediction Based on EEMD,\u0026rdquo; \u003cem\u003eEnergies\u003c/em\u003e, vol. 13, no. 9, 2020, doi: 10.3390/en13092380.\u003c/li\u003e\n\u003cli\u003eP. Shi, R. Hu, Y. He, X. Chen, H. Huang, and Z. Liu, \u0026ldquo;A novel LFSSA-LSTM algorithm for the remaining useful life prediction of lithium-ion batteries in whole life cycle,\u0026rdquo; \u003cem\u003eInt. J. Electrochem. Sci.\u003c/em\u003e, vol. 19, no. 9, p. 100719, Sep. 2024, doi: 10.1016/j.ijoes.2024.100719.\u003c/li\u003e\n\u003cli\u003eC. Lin, X. Tuo, L. Wu, G. Zhang, and X. Zeng, \u0026ldquo;Accurate Capacity Prediction and Evaluation with Advanced SSA-CNN-BiLSTM Framework for Lithium-Ion Batteries,\u0026rdquo; \u003cem\u003eBatteries\u003c/em\u003e, vol. 10, no. 3, Art. no. 3, Mar. 2024, doi: 10.3390/batteries10030071.\u003c/li\u003e\n\u003cli\u003eJ. Xia \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Historical data-independent remaining useful life prediction method based on dual-input deep learning neural network,\u0026rdquo; \u003cem\u003eJ. Energy Storage\u003c/em\u003e, vol. 72, p. 108427, Nov. 2023, doi: 10.1016/j.est.2023.108427.\u003c/li\u003e\n\u003cli\u003eP. Shi, R. Hu, Y. He, X. Chen, H. Huang, and Z. Liu, \u0026ldquo;A novel LFSSA-LSTM algorithm for the remaining useful life prediction of lithium-ion batteries in whole life cycle,\u0026rdquo; \u003cem\u003eInt. J. Electrochem. Sci.\u003c/em\u003e, vol. 19, no. 9, p. 100719, Sep. 2024, doi: 10.1016/j.ijoes.2024.100719.\u003c/li\u003e\n\u003cli\u003eC. Han and H. Yuan, \u0026ldquo;Capacity prediction of Lithium-ion batteries based on adaptive sliding window pooling extreme learning machine algorithm,\u0026rdquo; \u003cem\u003ePhys. Scr.\u003c/em\u003e, vol. 99, no. 1, p. 016003, Jan. 2024, doi: 10.1088/1402-4896/ad1704.\u003c/li\u003e\n\u003cli\u003eJ. Dou \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Extreme learning machine model for state-of-charge estimation of lithium-ion battery using salp swarm algorithm,\u0026rdquo; \u003cem\u003eJ. Energy Storage\u003c/em\u003e, vol. 52, p. 104996, Aug. 2022, doi: 10.1016/j.est.2022.104996.\u003c/li\u003e\n\u003cli\u003eY. Ji, Z. Chen, Y. Shen, K. Yang, Y. Wang, and J. Cui, \u0026ldquo;An RUL prediction approach for lithium-ion battery based on SADE-MESN,\u0026rdquo; \u003cem\u003eAppl Soft Comput\u003c/em\u003e, vol. 104, no. C, Jun. 2021, doi: 10.1016/j.asoc.2021.107195.\u003c/li\u003e\n\u003cli\u003eX. Bian, Z. Wei, J. He, F. Yan, and L. Liu, \u0026ldquo;A Novel Model-Based Voltage Construction Method for Robust State-of-Health Estimation of Lithium-Ion Batteries,\u0026rdquo; \u003cem\u003eIEEE Trans. Ind. Electron.\u003c/em\u003e, vol. 68, no. 12, pp. 12173\u0026ndash;12184, Dec. 2021, doi: 10.1109/TIE.2020.3044779.\u003c/li\u003e\n\u003cli\u003eX. Zhang, H. Li, W. Zhang, A. M. Lopes, X. Wu, and L. Chen, \u0026ldquo;Improved Parameter Identification for Lithium-Ion Batteries Based on Complex-Order Beetle Swarm Optimization Algorithm,\u0026rdquo; \u003cem\u003eMicromachines\u003c/em\u003e, vol. 14, no. 2, p. 413, Feb. 2023, doi: 10.3390/mi14020413.\u003c/li\u003e\n\u003cli\u003eT. Yuan, N. Li, H. Sun, and S. Tan, \u0026ldquo;A Joint Estimation Method of SOC and SOH for Lithium-ion Battery Considering Cyber-Attacks Based on GA-BP,\u0026rdquo; \u003cem\u003eComput. Mater. Contin.\u003c/em\u003e, vol. 80, no. 3, pp. 4497\u0026ndash;4512, 2024, doi: 10.32604/cmc.2024.056061.\u003c/li\u003e\n\u003cli\u003eS. Zhao, C. Zhang, and Y. Wang, \u0026ldquo;Lithium-ion battery capacity and remaining useful life prediction using board learning system and long short-term memory neural network,\u0026rdquo; \u003cem\u003eJ. Energy Storage\u003c/em\u003e, vol. 52, p. 104901, Aug. 2022, doi: 10.1016/j.est.2022.104901.\u003c/li\u003e\n\u003cli\u003eV. Safavi, A. Mohammadi Vaniar, N. Bazmohammadi, J. C. Vasquez, and J. M. Guerrero, \u0026ldquo;Battery Remaining Useful Life Prediction Using Machine Learning Models: A Comparative Study,\u0026rdquo; \u003cem\u003eInformation\u003c/em\u003e, vol. 15, no. 3, 2024, doi: 10.3390/info15030124.\u003c/li\u003e\n\u003cli\u003eH. Zhao, Z. Chen, X. Shu, J. Shen, Z. Lei, and Y. Zhang, \u0026ldquo;State of health estimation for lithium-ion batteries based on hybrid attention and deep learning,\u0026rdquo; \u003cem\u003eReliab. Eng. Syst. Saf.\u003c/em\u003e, vol. 232, p. 109066, Apr. 2023, doi: 10.1016/j.ress.2022.109066.\u003c/li\u003e\n\u003cli\u003eS. Jafari and Y.-C. Byun, \u0026ldquo;A CNN-GRU Approach to the Accurate Prediction of Batteries\u0026rsquo; Remaining Useful Life from Charging Profiles,\u0026rdquo; \u003cem\u003eComputers\u003c/em\u003e, vol. 12, no. 11, 2023, doi: 10.3390/computers12110219.\u003c/li\u003e\n\u003cli\u003eX. Qiao, Z. Wang, X. Su, G. Shen, E. Hou, and Y. Cai, \u0026ldquo;Hybrid approach for online capacity estimation of lithium-lon batteries: Integrating model-driven and data-driven methods,\u0026rdquo; \u003cem\u003eJ. Energy Storage\u003c/em\u003e, vol. 113, p. 115582, Mar. 2025, doi: 10.1016/j.est.2025.115582.\u003c/li\u003e\n\u003cli\u003eV. Yama\u0026ccedil;li, \u0026ldquo;State-of-health estimation and classification of series-connected batteries by using deep learning based hybrid decision approach,\u0026rdquo; \u003cem\u003eHeliyon\u003c/em\u003e, vol. 10, no. 20, p. e39121, Oct. 2024, doi: 10.1016/j.heliyon.2024.e39121.\u003c/li\u003e\n\u003cli\u003eAman and R. S. Chhillar, \u0026ldquo;Disease Prediction using Deep Learning Algorithms in Healthcare Sector,\u0026rdquo; in \u003cem\u003eDisease Prediction using Deep Learning Algorithms in Healthcare Sector\u003c/em\u003e, 108-115: Technoarete Publishing, 2022. doi: 10.36647/MLAIDA/2022.12.B1.Ch008.\u003c/li\u003e\n\u003cli\u003eA. Darolia, R. S. Chhillar, M. Alhussein, S. Dalal, K. Aurangzeb, and U. K. Lilhore, \u0026ldquo;Enhanced cardiovascular disease prediction through self-improved Aquila optimized feature selection in quantum neural network \u0026amp;amp; LSTM model,\u0026rdquo; \u003cem\u003eFront. Med.\u003c/em\u003e, vol. 11, Jun. 2024, doi: 10.3389/fmed.2024.1414637.\u003c/li\u003e\n\u003cli\u003eS. Jafari, J.-H. Yang, and Y.-C. Byun, \u0026ldquo;Optimized XGBoost modeling for accurate battery capacity degradation prediction,\u0026rdquo; \u003cem\u003eResults Eng.\u003c/em\u003e, vol. 24, p. 102786, Dec. 2024, doi: 10.1016/j.rineng.2024.102786.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Battery Degradation Prediction, Remaining Useful Life Estimation, Deep Learning in Energy Storage, Battery Cycle Life Prediction","lastPublishedDoi":"10.21203/rs.3.rs-6212719/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6212719/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAccurate prediction of battery degradation and remaining useful life (RUL) is critical for optimizing the performance and lifespan of battery-powered systems in electric vehicles and renewable energy storage applications. This paper introduces a novel machine learning approach utilizing Bidirectional Long Short-Term Memory (Bi-LSTM) networks to predict battery degradation and estimate RUL based on key parameters including voltage, current, temperature, and cycle count. Unlike conventional LSTM models that process data in a unidirectional manner, our Bi-LSTM architecture captures both past and future dependencies in battery behavior, significantly improving prediction accuracy. Through comprehensive evaluation on real-world battery datasets, we demonstrate that Bi-LSTM outperforms traditional LSTM systems by reducing root mean square error (RMSE) for state of health (SOH) prediction from 4.5\u0026ndash;3.1% and improving R\u0026sup2; values from 0.87 to 0.92. For RUL prediction, our model achieves an RMSE of 120 cycles compared to 150 cycles for standard LSTM. These improvements enable more reliable real-time battery health monitoring and proactive management strategies. The integration of Bi-LSTM into battery management systems (BMS) offers enhanced computational efficiency and superior convergence speed, making it particularly suitable for applications requiring precise battery management such as electric vehicles and grid-scale energy storage systems.\u003c/p\u003e","manuscriptTitle":"Enhanced Battery Degradation and RUL Prediction Using Bidirectional LSTM Networks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-14 06:44:02","doi":"10.21203/rs.3.rs-6212719/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2ae3ea14-f98f-4737-a45d-f338e1dfc451","owner":[],"postedDate":"March 14th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":45666820,"name":"Renewable Resources"},{"id":45666821,"name":"Electronic Materials and Devices"}],"tags":[],"updatedAt":"2025-03-14T06:44:02+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-14 06:44:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6212719","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6212719","identity":"rs-6212719","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}