A comprehensive review of machine learning prediction in the production of bio-oil from lignocellulose via pyrolysis

A comprehensive review of machine learning prediction in the production of bio-oil from lignocellulose via pyrolysis | 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 A comprehensive review of machine learning prediction in the production of bio-oil from lignocellulose via pyrolysis Hyojin Lee, Il-Ho Choi, Kyung-Ran Hwang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3830648/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 Bio-oil produced through pyrolysis of lignocellulosic biomass has recently received significant attention due to its possible uses as a second-generation biofuel. The yield and characteristics of produced bio-oil are affected by reaction conditions (reactor type, particle size, feed rate, operating temperature, heating rate, retention time, etc.) and the type of feedstock that is used (softwood, hardwood, agricultural plant residues, miscanthus, etc.). Recently, machine learning (ML) techniques have been widely employed to forecast the performance of the pyrolysis and the characteristics of bi-oil. In this study, a comprehensive review of ML research on bio-oil has been carried out. Regression methods were most frequently employed to build prediction models. The top five ML methods for bio-oil research were random forest, artificial neural network, gradient boosting, support vector regression, and linear regression. In addition, users frequently extract features using their own knowledge and restricted datasets were employed I previous studies. We highlighted the challenges and potential of cutting-edge ML techniques in bio-oil production. machine learning predictive modeling pyrolysis bio-oil lignocellulose Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction In the face of the climate crisis, we must strive in many ways to make a successful transition from today's fossil-based economy to a future circular bioeconomy. As part of this, the transportation sector should achieve carbon-neutral growth by expanding the supply and use of biofuels. Biofuels are a viable option, but there is not enough biomass available to meet the rapidly growing global demand for biofuels. Under these circumstances, many technologies are being developed and advanced for transition from 1st generation to 2nd and 3rd generation biofuels, especially lignocellulose-derived fuels through pyrolysis (Jahanshahi et al. 2023 ). Bio-oil, produced through pyrolysis of lignocellulosic biomass, has received strong interest due to its potential application. The bio-oil that meets the quality standards of ASTM D7544 or EN 16900 is directly used in industrial burners (Lehto et al. 2014 ). High-quality bio-jet fuel and advanced bio-diesel can be made from bio-oil through a multi-stage hydro-treating process including hydro-stabilization, hydro-deoxygenation, etc. (Gholizadeh et al. 2023 ). In addition, it can be used as a feedstock for the production of syngas or hydrogen via gasification (Zheng et al. 2019 ). However, the characteristics of bio-oil are greatly influenced by the type of biomass and the reaction conditions for pyrolysis (Abdullah et al. 2023 ). Moreover, bio-oil contains a large amount of moisture, oxygen, and carbonyl compounds, making it unstable and reducing its storage stability. Therefore, additional stabilization processes (such as alcohol mixing and mild hydrogen treating, etc.) and further deep hydro-deoxygenation processes are required to produce high-quality biofuels or for other uses. If the characteristics of bio-oil are inconsistent, a burden is placed on the subsequent chemical processes and a wider range of reaction conditions must be considered to accommodate bio-oil with various characteristics. This leads to an increase in the reaction stage, a greater amount of hydrogen, a change in the reaction temperature, etc., which make the process more difficult. To minimize this, it is important to predict the quality of bio-oil produced depending on the type of biomass and reaction conditions and also to set reaction (pyrolysis) conditions in advance according to the type of biomass to ensure consistent quality of bio-oil. As described above, the yield and composition distribution of bio-oil greatly depend on the following factors: the type of feedstock (softwood, hardwood, agricultural plant residues, miscanthus, etc.) and reaction conditions (reactor type, particle size, feed rate, operating temperature, heating rate, retention time, etc.) (Zhang and Matharu 2018 ; Guedes et al. 2018 ; Hafeez et al. 2019 ), as shown in Fig. 1 . This means that the bio-oil production process is a highly-interrelated, high dimensional system and its complexity inherently leads to considerable trial-and-error. Therefore, when intending to use a certain biomass or mixed biomass as a raw material, instead of labor- and time-consuming and costly experimental methods, predictive modeling is required in order to predict the quantity of bio-oil that can be obtained and its characteristics, or the reaction conditions that are necessary to obtain a specific bio-oil (Fig. 1 ). Physics-based modeling, the traditional modeling approach, requires a comprehensive understanding of the mechanism of the system and engineering principle. On the contrary, data-driven modeling requires a large enough and qualified data set and possibly could return counter-intuitive prediction results. Recent developments in data science have led to breakthroughs in Machine Learning (ML) techniques (Bradley et al. 2022 ). Therefore, ML has been gaining growing attention in the bio-oil research field. The objective of this paper is to provide a comprehensive review of ML applications for bio-oil research. First, after briefly explaining ML techniques to provide insight to enhance comprehension for useful applications, details of the state-of-the-art of the use of ML in bio-oil research will be discussed. Finally, we describe the future prospects of ML techniques in the production of promising bio-oil as a 2nd generation biofuel. 2. Overview of ML ML is a subset of Artificial Intelligence (AI) that includes deep learning (DL) (Fig. 2 ) (Janga et al. 2023 ). Today, ML is extensively used for data analysis, data-driven modeling, and decision-making areas. In particular, ML can quickly find patterns that could have been ignored by humans in data sets. This implies that ML can determine which variable is important, how important it is, and which input variables have a stronger correlation with a certain output variable. Moreover, data can be transformed into valuable predictions using ML and predictive analytics. The best system design, operating conditions, and production planning then can be suggested by an algorithm. In this section, an outline of ML is provided. ML creates mathematical models for prediction using training data. There are three types of ML: supervised learning (SL), unsupervised learning, and reinforcement learning (RL) (Lee et al. 2018 ). SL uses labeled data (X and Y) to learn relations between predictors and responses through classification (discrete response) and regression (continuous response). Unsupervised learning uses unlabeled data (X) to learn their distribution via clustering. RL learns input data (X) and critics (U) to discover the suitable action that maximizes reward. (Fig. 3 ) Each of the ML methods has specific applications for various problems. ML categories and algorithms are represented in Fig. 4 . SL algorithms predict output variables based on input variables, and are the most widely used type of ML for bio-oil research. Regression and classification algorithms are the two main subcategories of SL. Response type determines the type of the algorithms to use. The regression algorithms estimate continuous numerical values such as temperature, heating value, and yield, whereas the classification algorithms determine which category an observation falls into, such as chemical components. Regression algorithms find the best math function to match the training data. In addition, the regression analysis can provide particular details regarding a connection between multiple variables in addition to indicating whether it is significant. More precisely, it can provide the degree to which certain factors will influence a dependent variable. Regression methods can be roughly classified into three categories: polynomial, multiple linear, and linear. Linear regression (LR) is a fundamental form of regression in ML that involves a predictor and a dependent variable that are correlated in a linear way (Fig. 5 (a)). Concerning to the nonlinearity, multiple linear functions or a polynomial function can be used. Support vector machine (SVM) is one of the classification algorithms. Through small adjustments, this technique may also be used for regression analysis (Sharifzadeh et al. 2019 ). Support vector regression (SVR) finds the best fit of a hyperplane with an allowed error margin, ɛ-insensitive tube (Fig. 5 . (b)). Artificial neural networks (ANNs) are designed to mimic the structure of the human brain, allowing it to process and interpret vast amounts of data into information that can be put to use (Fig. 5 (c)). Therefore, ANNs are frequently utilized as they work well for complex and nonlinear systems (Sharifzadeh et al. 2019 ). Random Forest (RF) could be used for both classification and regression. RF uses multiple trees to predict the majority of the modes and the average prediction for clarification and regression problems, respectively (Fig. 5 (d)). Details of RF are explained well in 2019 by Fan et. al. (Xing et al. 2019b ). Gradient boosting (GB) can be used for both classification and regression. In boosting, several weak learners are created sequentially to enhance the overall performance of the method (Fig. 5 (e)). GB minimizes a loss function by training an ensemble of predictors one after the other, compensating for the mistakes of the preceding forecasters (Hastie et al. 2001 ). Unsupervised learning can be used for feature extraction. For example, Aghbashlo et al. collected a sludge pyrolysis dataset and applied principal component analysis to decrease dimensionality (Shahbeik et al. 2022 ). Supervised and unsupervised learning find hidden patterns of datasets, but they do not decide action. RL can decide optimal action considering uncertainty. RL allows sequential decision making and is used for various purposes such as control, scheduling, and planning (Lee et al. 2018 ). 3. ML applications in bio-oil research In the last few years, various ML algorithms have been investigated to predict the performance of pyrolysis. The primary objective of previous research is to predict bio-oil yield, contents, heating values based on contents of biomass, and operating conditions (Table 1 ). The top five ML algorithms used for modelling of bio-oil production are as follows: RF, ANN, GB, SVR, and LR. Table 1 The top five ML algorithms for modelling of bio-oil production Algorism Input variables Output variables (Prediction targets) Tested method Best method Performance Ref. LR •Ash •Volatiles •Feedstock O/C ratio •Yield LR LR (Oasmaa et al. 2010 ) •Gas chromatography –mass spectrometry •HHV LR LR (Wanignon Ferdinand et al. 2012 ) •13C NMR •Mass fractions of C, H, N and O •HHV •Phenol & cresols conc. • Total acid number PLS PLS R 2 > 0.65 (Strahan et al. 2016 ) RF •Sludge ultimate composition •Proximate compositions •Temperature •Heating rate •Reaction time •Bio-oil yield •Syngas yield •Syngas composition •Biochar yield •Biochar atomic composition •Biochar calorific value MLPNN, SVR, RFR RF R 2 > 0.813 (Shahbeik et al. 2022 ) •Feedstock composition •Operating conditions •Yield •Hydrogen content MLR, RF RF R 2 = 0.92 (Tang et al. 2020 ) •Biomass compositions •Operating conditions •Yield •Viscosity, HHV •O/C and H/C ratio RF RF R 2 > 0.75 (Zhang et al. 2022 ) ANN •Temperature •Heating rate •Bio-char yield •Bio-oil yield ANN, RSM ANN R 2 > 0.922 (Angın and Tiryaki 2016 ) •Feedstock composition •Catalyst type and ratio •Operating conditions •Bio-oil yield ANN ANN R 2 > 0.933 (ÖZBAY and KÖKTEN 2020 ) •Temperature •Heating rate •Conversion •Activation energy ANN ANN R 2 = 0.999 (Asghar et al. 2023 ) •Operating conditions •Yield ANN, GNN GNN R 2 = 0.9924 (Singh et al. 2023 ) SVR •Feedstock composition •Operating conditions •Three-phase product distribution •HHV ANN, SVM SVM R 2 = 0.89 (Chen et al. 2018 ) •Feedstock characterization •Operating conditions •Yield SVR R 2 > 0.91 (Potnuri et al. 2023 ) GB •Biomass compositions •Operating conditions •Bio-oil yield •Biochar yield XGB, DNN XGB R 2 = 0.96 (Alabdrabalnabi et al. 2022 ) •Biomass compositions •Microwave power •Temperature •Yield •Product contents, HHV SVR, RF, GBR GBR R 2 > 0.823 (Yang et al. 2022 ) •Feedstock type •Feedstock composition •Temperature /heating rate •Char, oil, gas yield ANN, GBR, DT, RF, KNN, BR GBR/BR R 2 > 0.9 (Shen et al. 2022 ) •Biomass compositions •Operating conditions •Yield •Nitrogen heterocycles RF, GBR RF R 2 > 0.80 (Leng et al. 2023 ) In the early stage, simple LR was used for bio-oil yield prediction. Oasmaa et al. used a multiple linear regression model to explore the correlation between bio-oil yield and feedstock characteristics (ash, volatiles, and O/C ratio) (Oasmaa et al. 2010 ). Ferdinand et al. predicted higher heating value (HHV) of bio-oil based on GC–MS analysis data by using a conventional multiple regression model (Wanignon Ferdinand et al. 2012 ). Mullen et al. used the partial least-squares (PLS) regression model to predict the HHV of bio-oil based on the 13C NMR analysis information (Strahan et al. 2016 ). Rao et al. used a polynomial regression algorithm to estimate the impact of catalyst quantity and pre-treatment temperature on yield, heating rate, pyrolysis temperature, and conversion efficiency (Potnuri et al. 2022 ). Meanwhile, more advanced method were used for predicting the yield of biochar (Cao et al. 2016 ), gas production selectivity (Sun et al. 2016 ), and bio-diesel production (Mostafaei et al. 2016 ). To the best of our knowledge, in 2016, Angin et al. became the first group to apply the ANN method to predict the yield of bio-oil produced through pyrolysis and they concluded that the ANN model can replace the response surface methodology due to its high accuracy and generalization capability (Angın and Tiryaki 2016 ). ML studies for the prediction of bio-oil production and its characteristics via thermochemical reaction have been actively underway since 2020, as shown in Fig. S1 . Ozbay et al. developed an ANN model to predict the bio-oil yield based on proximate and ultimate analyses of biomass, catalyst type, and operating conditions. The slow and intermediate pyrolysis data were used and the prediction results were quite consistent with experiment results (ÖZBAY and KÖKTEN 2020 ). Mehmood et al. employed a multilayer perceptron-based ANN regression model for determining the activation energy of Saccharum Bengalense pyrolysis (Asghar et al. 2023 ). Singh et al. applied an ANN and generalized neural network (GNN) to predict bio-oil yield from lychee-based biomass based on pyrolysis parameters such as temperature, gas flow, heating rate, and retention time. GNN offered more precision in comparison to ANN (Singh et al. 2023 ). SVR was also often utilized to forecast product attributes and yield. Yuan et al. forecasted the biochar yield from cattle manure utilizing a least-squares support vector machine and an ANN based on 33 experimental datasets (Cao et al. 2016 ). Xiao et al. developed prediction models for the three-phase product distribution and bio-oil heating value using an ANN and SVM. They reported that SVR showed better performance (Chen et al. 2018 ). Rao et al. successfully utilized a SVM to predict the product yields from the co-pyrolysis of biomass and plastics. This study considered the catalyst and blend as input features (Potnuri et al. 2023 ). The RF method is the most often used ML algorithm for biomass research. It was mainly applied to predict the composition contents (cellulose, hemicellulose, and lignin) (Xing et al. 2019b ) and HHV (Xing et al. 2019a ) of biomass and to forecast the yield of bio-char produced by pyrolysis (Zhu et al. 2019 ). Yang et al. used the RF method for predicting bio-oil yield and hydrogen content of bio-oil, based on the types of biomass and pyrolysis reaction conditions. Feedstock compositions had a greater an impact on both yield and hydrogen content (Tang et al. 2020 ). In 2021, Yang et al. used the RF method to predict gas yield and its major composition. They reported that pyrolysis conditions contributed more to predict yield and H 2 and CO 2 concentration (Tang et al. 2021 ). Wang et al. used the RF method to develop prediction models for predicting the yield and carbon content of biochar based on the pyrolysis data of lignocellulosic biomass. The yield and C-char changes were primarily attributed to the pyrolysis temperature (Zhu et al. 2019 ). Mu et al. used RF to predict yield, HHV, viscosity, and oxygen and carbon ratio of bio-oil. They reported that pyrolysis conditions (temperature, heating rate, and particle size) is less influential than the ultimate and proximate analysis of feedstock (Zhang et al. 2022 ). In previous studies, the most important input characteristic differed depending on the dataset. The variety of feedstocks and pyrolysis conditions leads to an inconsistent input/output correlation, and this highlights the significance of the dataset. Recently, GB methods were frequently used. Gautam et al. developed a dense neural network and the XGBoost (XGB) model to predict the yield of co-pyrolysis of biomass and waste plastic. XGB offered the best performance with a root mean squared error of 1.77 and an R 2 of 0.96 (Alabdrabalnabi et al. 2022 ). Tabatabaei compared SVR, RF, and gradient boosting regressor (GBR) for microwave-assisted pyrolysis. Among them, GBR provides better prediction performance with a R 2 > 0.822 (Yang et al. 2022 ). Gao et al. used ANN, GB, decision trees (DT), RF, K-nearest-neighbors, bagging regressor, and lasso regression to predict char, bio-oil, and gas product yields. GB provided better predictions with a R 2 > 0.90 in single output models, whereas the bagging regressor showed the best performance in multiple output models (Shen et al. 2022 ). Wu et al. developed the machine-learning assisted Aspen Plus rector model, which predicts the yield, HHV, and yield component distribution of biocrude oil according to the experimental records of the hydrothermal liquefaction process from 16 species of microalgae. The least-square boosting model showed the best performance for the prediction (Wu et al. 2023 ). Yang et al. compared RF and GBR in terms of prediction of nitrogen heterocycles of bio-oil. For this, 217 datasets were collected from 63 SCI papers comprising 91 biomasses and the Pearson correlation coefficient was used. In this study, RF returned better prediction than GBR for every case (Leng et al. 2023 ). Various ML methods will be used based on the advancement of the ML algorithm. This will be discussed in the next section. Additionally, Colosi et al. made further improvement by integrating ML, life cycle assessment (LCA), and techno-economic analysis (TEA). The RF model predicts the yield and characteristics of bio-char, and calculates the parameters of LCA and TEA such as energy return on investment, global warming potential, and minimum product selling price (Cheng et al. 2020 ). Chemmangattuvalappil et al. developed a rule-based model using rough set machine learning (RSML), which allows to combine expert knowledge and a dataset. In this study, 207 and 128 data points in published studies were used to estimate the HHV and pH of bio-oil, respectively (Chong et al. 2022 ). Xiong et al. used historical computational fluid dynamics (CFD) data to build a yield prediction model. The long and short term memory network was used to predict the mass flow rates at the reactor. The CFD simulation time was reduced by almost 30% through the application of ML (Zhong et al. 2023 ). Khan et al. applied five different ML methods (SVM, ANN, DT, Gaussian process regression, and ensembled tree) integrated with two optimization methods (particle swarm optimization and genetic algorithm) for feature selection and hyperparameter optimization of ML models. A Gaussian process regression - genetic algorithm model showed the best performance (Ullah et al. 2023 ). Baskar et al investigated role of AI technologies in the municipal solid waste management area. They showed that AI can play a significant role not only for the prediction of reaction parameters but also the sorting process (Naveenkumar et al. 2023 ). In summary, previous studies mainly used ML to predict bio-oil and gas yield and the component and characteristics of products based on feedstock type, components, operating conditions, and pyrolysis type. Therefore, most ML applications have been regressions. Frequently, users extract features using their own knowledge. This also means that the datasets that were used were limited. The limitations of previous studies include use of confined data and limited use of ML algorithms for the prediction of the pyrolysis reaction. 4. Future Outlook 4.1. Dataset and DL Bio-oil can be produced through various pyrolysis conditions and feedstocks. Numerous bio-oil datasets, produced through both experimentation and modeling/simulation, are obtainable in the literature. However, in previous studies, most datasets learned by ML were collected by users with specific goals in mind. In other words, the confined datasets were used for specific purposes, and this resulted in the usage of constrained algorithms, namely regression. For ML applications, data collection is always essential. Both the quality and quantity of data are important. Learning a sufficient number of qualified datasets by ML may provide new insight, including possibly even counter-intuitive findings. If there are enough labeled datasets, then DL techniques can be used. DL can tackle complex problems even when the datasets are exceedingly diverse and unstructured (Taye 2023 ). It is expected that DL would be advantageous to the bio-oil research community. 4.2. Hybrid model In previous studies, researchers utilized their knowledge only for problem formulation (i.e. define input/output variables), and not problem solving. In order to address this deficiency, Chemmangattuvalappil et al. adapted the RSML approach (Chong et al. 2022 ). Rule-based model ML approaches enable the merging of expert knowledge with data incorporated in information throughout the training process. Namely, domain knowledge may be added through user-specified training parameters to guarantee that the final rules make sense in terms of physical mechanisms. The black-box ML techniques have low intrinsic interpretability, poor extrapolating capabilities, and unbounded uncertainty in predictions. Attempts to address the deficiency include a hybrid model that learns from data and physics. In other words, combining ML with first-principles knowledge is the next horizon since it can increase accuracy and interpretability while utilizing less data. The hybrid concept was well articulated by Boukouvala (Bradley et al. 2022 ). 4.3. Optimal decision making: process, control, scheduling, design of experiment ML is capable of making optimal decisions. In particular, RL learns the optimal decision policy and links the system state to the optimal course of action. As decisions made at one time have an impact on subsequent decisions and the resulting output events, this is a natural feature of a dynamical system; this is where RL differs from SL. RL uses evaluative feedback from the environment to estimate real-valued rewards or costs, whereas SL uses informative feedback through classification errors (Lee et al. 2018 ). Therefore, ML may be applied to optimal control, planning, and scheduling. Additionally, in terms of design of experiment, Bayesian optimization (BO) is a useful tool. It has been shown that BO is sample-efficient and scalable, requiring minimal testing and enabling the exploration of large design spaces (González and Zavala 2023 ). BO can be employed in bio-oil pyrolysis experiments to reduce trial and error. 5. Conclusion A comprehensive review of bio-oil pyrolysis research using ML has been performed. In the last decade, numerous ML approaches have been studied to forecast pyrolysis performance. Target variables were mainly bio-oil and gas yield, contents, and chemical properties of products, and predictors were feedstock type, ultimate and proximate composition of feedstock, reactor type, and operating conditions such as temperature and retention time. The most frequently used ML was regression algorithms: RF, ANN, GB, SVR, and LR. The limitations of previous studies are use of confined data and limited use of ML algorithms for prediction of the pyrolysis reaction. We anticipate that ML will aid the bio-oil research field in further utilizing DL, hybrid models, RL, and BO. Abbreviations ML Machine Learning AI Artificial Intelligence DL Deep Learning SL Supervised Learning RL Reinforcement Learning LR Linear Regression SVM Support Vector Machine SVR Support Vector Regression ANNs Artificial Neural Networks RF Random Forest GB Gradient Boosting HHV Higher Heating Value PLS Partial Least-Squares GNN Generalized Neural Network XGB XGBoost GBR Gradient Boosting Regressor DT Decision Tree LCA Life Cycle Assessment TEA Techno-Economic Analysis RSML Rough Set Machine Learning CFD Computational Fluid Dynamics BO Bayesian Optimization Declarations CRediT authorship contribution statement Hyojin Lee : Conceptualization, Methodology, Investigation, Visualization, Writing – original draft. IL-Ho Choi : Visualization, Writing – review & editing. Kyung-Ran Hwang : Conceptualization, Supervision, Project administration, Writing – review & editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could appear to influence the work reported in this paper. Data availability No data were used for the research described in the article. Acknowledgements This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No.NRF2020M1A2A2079802). Appendix A. Supplementary data Supplementary data for this work can be found in the e-version of this paper online. References Abdullah N, Mohd Taib R, Mohamad Aziz NS, et al (2023) Banana pseudo-stem biochar derived from slow and fast pyrolysis process. Heliyon 9:e12940. https://doi.org/10.1016/j.heliyon.2023.e12940 Alabdrabalnabi A, Gautam R, Mani Sarathy S (2022) Machine learning to predict biochar and bio-oil yields from co-pyrolysis of biomass and plastics. Fuel 328:125303. https://doi.org/10.1016/j.fuel.2022.125303 Angın D, Tiryaki AE (2016) Application of response surface methodology and artificial neural network on pyrolysis of safflower seed press cake. Energy Sources, Part A: Recovery, Utilization, and Environmental Effects 38:1055–1061. https://doi.org/10.1080/15567036.2013.862585 Asghar A, Liu C-G, Ali I, et al (2023) Bioenergy potential of Saccharum bengalense through pyrolysis, reaction kinetics, TG-FTIR-GCMS analysis of pyrolysis products, and validation of the pyrolysis data through machine learning. Chemical Engineering Journal 465:142930. https://doi.org/10.1016/j.cej.2023.142930 Bradley W, Kim J, Kilwein Z, et al (2022) Perspectives on the integration between first-principles and data-driven modeling. Comput Chem Eng 166:107898. https://doi.org/10.1016/j.compchemeng.2022.107898 Cao H, Xin Y, Yuan Q (2016) Prediction of biochar yield from cattle manure pyrolysis via least squares support vector machine intelligent approach. Bioresour Technol 202:158–164. https://doi.org/10.1016/j.biortech.2015.12.024 Cheng F, Luo H, Colosi LM (2020) Slow pyrolysis as a platform for negative emissions technology: An integration of machine learning models, life cycle assessment, and economic analysis. Energy Convers Manag 223:113258. https://doi.org/10.1016/j.enconman.2020.113258 Chen X, Zhang H, Song Y, Xiao R (2018) Prediction of product distribution and bio-oil heating value of biomass fast pyrolysis. Chemical Engineering and Processing - Process Intensification 130:36–42. https://doi.org/10.1016/j.cep.2018.05.018 Chong JW, Thangalazhy‐Gopakumar S, Tan RR, et al (2022) Estimation of fast pyrolysis bio‐oil properties from feedstock characteristics using rough‐set ‐based machine learning. Int J Energy Res 46:19159–19176. https://doi.org/10.1002/er.8201 Gholizadeh M, Zhang S, Hu X, Wang Y (2023) Advances and Perspectives of Bio-oil Hydrotreatment for Biofuel Production. Energy & Fuels 37:10134–10154. https://doi.org/10.1021/acs.energyfuels.3c01647 González LD, Zavala VM (2023) New paradigms for exploiting parallel experiments in Bayesian optimization. Comput Chem Eng 170:108110. https://doi.org/10.1016/j.compchemeng.2022.108110 Guedes RE, Luna AS, Torres AR (2018) Operating parameters for bio-oil production in biomass pyrolysis: A review. J Anal Appl Pyrolysis 129:134–149. https://doi.org/10.1016/j.jaap.2017.11.019 Hafeez S, Pallari E, Manos G, Constantinou A (2019) Catalytic Conversion and Chemical Recovery. In: Plastics to Energy. Elsevier, pp 147–172 Hastie T, Friedman J, Tibshirani R (2001) The Elements of Statistical Learning. Springer New York, New York, NY Jahanshahi A, Lopes M, Brandão M, De Castro EA (2023) Development of bioenergy technologies: A scientometric analysis. Heliyon 9:e20000. https://doi.org/10.1016/j.heliyon.2023.e20000 Janga JK, Reddy KR, Raviteja KVNS (2023) Integrating artificial intelligence, machine learning, and deep learning approaches into remediation of contaminated sites: A review. Chemosphere 345:140476. https://doi.org/10.1016/j.chemosphere.2023.140476 Lee JH, Shin J, Realff MJ (2018) Machine learning: Overview of the recent progresses and implications for the process systems engineering field. Comput Chem Eng 114:111–121. https://doi.org/10.1016/j.compchemeng.2017.10.008 Lehto J, Oasmaa A, Solantausta Y, et al (2014) Review of fuel oil quality and combustion of fast pyrolysis bio-oils from lignocellulosic biomass. Appl Energy 116:178–190. https://doi.org/10.1016/j.apenergy.2013.11.040 Leng L, Li T, Zhan H, et al (2023) Machine learning-aided prediction of nitrogen heterocycles in bio-oil from the pyrolysis of biomass. Energy 278:127967. https://doi.org/10.1016/j.energy.2023.127967 Mostafaei M, Javadikia H, Naderloo L (2016) Modeling the effects of ultrasound power and reactor dimension on the biodiesel production yield: Comparison of prediction abilities between response surface methodology (RSM) and adaptive neuro-fuzzy inference system (ANFIS). Energy 115:626–636. https://doi.org/10.1016/j.energy.2016.09.028 Naveenkumar R, Iyyappan J, Pravin R, et al (2023) A strategic review on sustainable approaches in municipal solid waste management and energy recovery: Role of artificial intelligence, economic stability and life cycle assessment. Bioresour Technol 379:129044. https://doi.org/10.1016/j.biortech.2023.129044 Oasmaa A, Solantausta Y, Arpiainen V, et al (2010) Fast Pyrolysis Bio-Oils from Wood and Agricultural Residues. Energy & Fuels 24:1380–1388. https://doi.org/10.1021/ef901107f ÖZBAY G, KÖKTEN ES (2020) Modeling of Bio-Oil Production by Pyrolysis of Woody Biomass: Artificial Neural Network Approach. Politeknik Dergisi 23:1255–1264. https://doi.org/10.2339/politeknik.659136 Potnuri R, Rao CS, Surya DV, et al (2023) Utilizing support vector regression modeling to predict pyro product yields from microwave-assisted catalytic co-pyrolysis of biomass and waste plastics. Energy Convers Manag 292:117387. https://doi.org/10.1016/j.enconman.2023.117387 Potnuri R, Suriapparao D V., Sankar Rao C, et al (2022) Effect of dry torrefaction pretreatment of the microwave-assisted catalytic pyrolysis of biomass using the machine learning approach. Renew Energy 197:798–809. https://doi.org/10.1016/j.renene.2022.08.006 Shahbeik H, Rafiee S, Shafizadeh A, et al (2022) Characterizing sludge pyrolysis by machine learning: Towards sustainable bioenergy production from wastes. Renew Energy 199:1078–1092. https://doi.org/10.1016/j.renene.2022.09.022 Sharifzadeh M, Sikinioti-Lock A, Shah N (2019) Machine-learning methods for integrated renewable power generation: A comparative study of artificial neural networks, support vector regression, and Gaussian Process Regression. Renewable and Sustainable Energy Reviews 108:513–538. https://doi.org/10.1016/j.rser.2019.03.040 Shen J, Yan M, Fang M, Gao X (2022) Machine learning-based modeling approaches for estimating pyrolysis products of varied biomass and operating conditions. Bioresour Technol Rep 20:101285. https://doi.org/10.1016/j.biteb.2022.101285 Singh Y, Singh D, Singh NK, et al (2023) Production of bio-oil from lychee-based biomass through pyrolysis and maximization of bio-oil yield with statistical and machine learning techniques. J Clean Prod 413:137472. https://doi.org/10.1016/j.jclepro.2023.137472 Strahan GD, Mullen CA, Boateng AA (2016) Prediction of Properties and Elemental Composition of Biomass Pyrolysis Oils by NMR and Partial Least Squares Analysis. Energy & Fuels 30:423–433. https://doi.org/10.1021/acs.energyfuels.5b02345 Sun Y, Liu L, Wang Q, et al (2016) Pyrolysis products from industrial waste biomass based on a neural network model. J Anal Appl Pyrolysis 120:94–102. https://doi.org/10.1016/j.jaap.2016.04.013 Tang Q, Chen Y, Yang H, et al (2020) Prediction of Bio-oil Yield and Hydrogen Contents Based on Machine Learning Method: Effect of Biomass Compositions and Pyrolysis Conditions. Energy & Fuels 34:11050–11060. https://doi.org/10.1021/acs.energyfuels.0c01893 Tang Q, Chen Y, Yang H, et al (2021) Machine learning prediction of pyrolytic gas yield and compositions with feature reduction methods: Effects of pyrolysis conditions and biomass characteristics. Bioresour Technol 339:125581. https://doi.org/10.1016/j.biortech.2021.125581 Taye MM (2023) Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Future Directions. Computers 12:91. https://doi.org/10.3390/computers12050091 Ullah H, Haq ZU, Naqvi SR, et al (2023) Optimization based comparative study of machine learning methods for the prediction of bio-oil produced from microalgae via pyrolysis. J Anal Appl Pyrolysis 170:105879. https://doi.org/10.1016/j.jaap.2023.105879 Wanignon Ferdinand F, Van de Steene L, Kamenan Blaise K, Siaka T (2012) Prediction of pyrolysis oils higher heating value with gas chromatography–mass spectrometry. Fuel 96:141–145. https://doi.org/10.1016/j.fuel.2012.01.007 Wu W, Huang C-M, Tsai Y-H (2023) Design and validation of a microalgae biorefinery using machine learning-assisted modeling of hydrothermal liquefaction. Algal Res 74:103230. https://doi.org/10.1016/j.algal.2023.103230 Xing J, Luo K, Wang H, et al (2019a) A comprehensive study on estimating higher heating value of biomass from proximate and ultimate analysis with machine learning approaches. Energy 188:116077. https://doi.org/10.1016/j.energy.2019.116077 Xing J, Luo K, Wang H, Fan J (2019b) Estimating biomass major chemical constituents from ultimate analysis using a random forest model. Bioresour Technol 288:121541. https://doi.org/10.1016/j.biortech.2019.121541 Yang Y, Shahbeik H, Shafizadeh A, et al (2022) Biomass microwave pyrolysis characterization by machine learning for sustainable rural biorefineries. Renew Energy 201:70–86. https://doi.org/10.1016/j.renene.2022.11.028 Zhang T, Cao D, Feng X, et al (2022) Machine learning prediction of bio-oil characteristics quantitatively relating to biomass compositions and pyrolysis conditions. Fuel 312:122812. https://doi.org/10.1016/j.fuel.2021.122812 Zhang Z, Matharu AS (2018) Thermochemical Valorization of Paper Deinking Residue through Microwave-Assisted Pyrolysis. In: Waste Biorefinery. Elsevier, pp 671–692 Zheng J-L, Zhu Y-H, Zhu M-Q, et al (2019) A review of gasification of bio-oil for gas production. Sustain Energy Fuels 3:1600–1622. https://doi.org/10.1039/C8SE00553B Zhong H, Wei Z, Man Y, et al (2023) Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data. J Clean Prod 391:136192. https://doi.org/10.1016/j.jclepro.2023.136192 Zhu X, Li Y, Wang X (2019) Machine learning prediction of biochar yield and carbon contents in biochar based on biomass characteristics and pyrolysis conditions. Bioresour Technol 288:121527. https://doi.org/10.1016/j.biortech.2019.121527 Additional Declarations No competing interests reported. Supplementary Files Appendics.docx 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-3830648","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":264937425,"identity":"a14fea21-dfe8-40ce-b11c-ef09bf6e0ecc","order_by":0,"name":"Hyojin Lee","email":"","orcid":"","institution":"Korea Institute of Energy Research","correspondingAuthor":false,"prefix":"","firstName":"Hyojin","middleName":"","lastName":"Lee","suffix":""},{"id":264937426,"identity":"17ddb29c-af05-484e-b553-b8d765f52083","order_by":1,"name":"Il-Ho Choi","email":"","orcid":"","institution":"Korea Institute of Energy Research","correspondingAuthor":false,"prefix":"","firstName":"Il-Ho","middleName":"","lastName":"Choi","suffix":""},{"id":264937427,"identity":"b002c9a1-f602-4484-82b8-7efaf06934e0","order_by":2,"name":"Kyung-Ran Hwang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAu0lEQVRIiWNgGAWjYBACAx4QWZHAA+UnEKvlDMlaGNvgKonQYs5z9uHnwnlpMvzSB9g+fGBIyyeoxbK33Vh65rYcHsm+BOaZMxhyLBsIOuw8G4M077YKHoMzDMzMPAwVBgRtAWph/s07p4LHHqTlD1FazraxSfM25PAAw4GZmYEhhwgtZ46xWfMcS+OROMPYzNhjkEaMljTm2zw1yfb8PcyHGX5UJBPWggQYG4AmkKJhFIyCUTAKRgFOAABffC6+Z+W8TwAAAABJRU5ErkJggg==","orcid":"","institution":"Korea Institute of Energy Research","correspondingAuthor":true,"prefix":"","firstName":"Kyung-Ran","middleName":"","lastName":"Hwang","suffix":""}],"badges":[],"createdAt":"2024-01-03 00:59:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3830648/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3830648/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49170254,"identity":"d60870b9-ce33-4189-a0c5-2475e0456e9d","added_by":"auto","created_at":"2024-01-04 10:03:20","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":99137,"visible":true,"origin":"","legend":"\u003cp\u003eBio-oil production via ML predictive models\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/878e23eb5023537b9b733c0c.png"},{"id":49170567,"identity":"1ac6a16f-5798-4899-aaf5-875867a69089","added_by":"auto","created_at":"2024-01-04 10:11:20","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":48828,"visible":true,"origin":"","legend":"\u003cp\u003eAI, ML, and DL\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/2920bb48df4b57bed5985a3e.png"},{"id":49170568,"identity":"acce86b8-d92b-4760-8cda-faa09e84ad6b","added_by":"auto","created_at":"2024-01-04 10:11:20","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":44671,"visible":true,"origin":"","legend":"\u003cp\u003eThree types of ML (a) SL (b) Unsupervised Learning (c) RL\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/7155ff095d66fbbe16fdb2e9.png"},{"id":49170257,"identity":"59b7a3e4-a3dc-4675-8797-0fa4a56428b3","added_by":"auto","created_at":"2024-01-04 10:03:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":66908,"visible":true,"origin":"","legend":"\u003cp\u003eML Algorithms\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/c959a84fab7d51797e15ecba.png"},{"id":49170258,"identity":"adb40b47-c056-49a2-a6ef-01256ffe13ac","added_by":"auto","created_at":"2024-01-04 10:03:20","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":69184,"visible":true,"origin":"","legend":"\u003cp\u003eConcept diagram of ML (a) LR (b) SVR (c) ANN (d) RF (e) GB\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/a9c8e884e993c32f2c3f4b69.png"},{"id":49505096,"identity":"68c1298b-5a0d-4272-b712-51976de94473","added_by":"auto","created_at":"2024-01-12 03:07:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":567919,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/578bfb49-a3fb-47c9-be6c-e6f6fb40aaaf.pdf"},{"id":49170993,"identity":"81f1a5cf-117a-435c-9527-26cc6be7b04a","added_by":"auto","created_at":"2024-01-04 10:19:20","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":29136,"visible":true,"origin":"","legend":"","description":"","filename":"Appendics.docx","url":"https://assets-eu.researchsquare.com/files/rs-3830648/v1/d484740fa48ae0b0657bf7f6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"A comprehensive review of machine learning prediction in the production of bio-oil from lignocellulose via pyrolysis","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIn the face of the climate crisis, we must strive in many ways to make a successful transition from today's fossil-based economy to a future circular bioeconomy. As part of this, the transportation sector should achieve carbon-neutral growth by expanding the supply and use of biofuels. Biofuels are a viable option, but there is not enough biomass available to meet the rapidly growing global demand for biofuels. Under these circumstances, many technologies are being developed and advanced for transition from 1st generation to 2nd and 3rd generation biofuels, especially lignocellulose-derived fuels through pyrolysis (Jahanshahi et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBio-oil, produced through pyrolysis of lignocellulosic biomass, has received strong interest due to its potential application. The bio-oil that meets the quality standards of ASTM D7544 or EN 16900 is directly used in industrial burners (Lehto et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). High-quality bio-jet fuel and advanced bio-diesel can be made from bio-oil through a multi-stage hydro-treating process including hydro-stabilization, hydro-deoxygenation, etc. (Gholizadeh et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In addition, it can be used as a feedstock for the production of syngas or hydrogen via gasification (Zheng et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). However, the characteristics of bio-oil are greatly influenced by the type of biomass and the reaction conditions for pyrolysis (Abdullah et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Moreover, bio-oil contains a large amount of moisture, oxygen, and carbonyl compounds, making it unstable and reducing its storage stability. Therefore, additional stabilization processes (such as alcohol mixing and mild hydrogen treating, etc.) and further deep hydro-deoxygenation processes are required to produce high-quality biofuels or for other uses. If the characteristics of bio-oil are inconsistent, a burden is placed on the subsequent chemical processes and a wider range of reaction conditions must be considered to accommodate bio-oil with various characteristics. This leads to an increase in the reaction stage, a greater amount of hydrogen, a change in the reaction temperature, etc., which make the process more difficult. To minimize this, it is important to predict the quality of bio-oil produced depending on the type of biomass and reaction conditions and also to set reaction (pyrolysis) conditions in advance according to the type of biomass to ensure consistent quality of bio-oil.\u003c/p\u003e \u003cp\u003eAs described above, the yield and composition distribution of bio-oil greatly depend on the following factors: the type of feedstock (softwood, hardwood, agricultural plant residues, miscanthus, etc.) and reaction conditions (reactor type, particle size, feed rate, operating temperature, heating rate, retention time, etc.) (Zhang and Matharu \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Guedes et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Hafeez et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. This means that the bio-oil production process is a highly-interrelated, high dimensional system and its complexity inherently leads to considerable trial-and-error. Therefore, when intending to use a certain biomass or mixed biomass as a raw material, instead of labor- and time-consuming and costly experimental methods, predictive modeling is required in order to predict the quantity of bio-oil that can be obtained and its characteristics, or the reaction conditions that are necessary to obtain a specific bio-oil (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Physics-based modeling, the traditional modeling approach, requires a comprehensive understanding of the mechanism of the system and engineering principle. On the contrary, data-driven modeling requires a large enough and qualified data set and possibly could return counter-intuitive prediction results. Recent developments in data science have led to breakthroughs in Machine Learning (ML) techniques (Bradley et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Therefore, ML has been gaining growing attention in the bio-oil research field.\u003c/p\u003e \u003cp\u003eThe objective of this paper is to provide a comprehensive review of ML applications for bio-oil research. First, after briefly explaining ML techniques to provide insight to enhance comprehension for useful applications, details of the state-of-the-art of the use of ML in bio-oil research will be discussed. Finally, we describe the future prospects of ML techniques in the production of promising bio-oil as a 2nd generation biofuel.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"2. Overview of ML","content":"\u003cp\u003eML is a subset of Artificial Intelligence (AI) that includes deep learning (DL) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) (Janga et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eToday, ML is extensively used for data analysis, data-driven modeling, and decision-making areas. In particular, ML can quickly find patterns that could have been ignored by humans in data sets. This implies that ML can determine which variable is important, how important it is, and which input variables have a stronger correlation with a certain output variable. Moreover, data can be transformed into valuable predictions using ML and predictive analytics. The best system design, operating conditions, and production planning then can be suggested by an algorithm. In this section, an outline of ML is provided.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eML creates mathematical models for prediction using training data. There are three types of ML: supervised learning (SL), unsupervised learning, and reinforcement learning (RL) (Lee et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). SL uses labeled data (X and Y) to learn relations between predictors and responses through classification (discrete response) and regression (continuous response). Unsupervised learning uses unlabeled data (X) to learn their distribution via clustering. RL learns input data (X) and critics (U) to discover the suitable action that maximizes reward. (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e) Each of the ML methods has specific applications for various problems.\u003c/p\u003e\u003cp\u003eML categories and algorithms are represented in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. SL algorithms predict output variables based on input variables, and are the most widely used type of ML for bio-oil research. Regression and classification algorithms are the two main subcategories of SL. Response type determines the type of the algorithms to use. The regression algorithms estimate continuous numerical values such as temperature, heating value, and yield, whereas the classification algorithms determine which category an observation falls into, such as chemical components. Regression algorithms find the best math function to match the training data. In addition, the regression analysis can provide particular details regarding a connection between multiple variables in addition to indicating whether it is significant. More precisely, it can provide the degree to which certain factors will influence a dependent variable.\u003c/p\u003e \u003cp\u003eRegression methods can be roughly classified into three categories: polynomial, multiple linear, and linear. Linear regression (LR) is a fundamental form of regression in ML that involves a predictor and a dependent variable that are correlated in a linear way (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(a)). Concerning to the nonlinearity, multiple linear functions or a polynomial function can be used. Support vector machine (SVM) is one of the classification algorithms. Through small adjustments, this technique may also be used for regression analysis (Sharifzadeh et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Support vector regression (SVR) finds the best fit of a hyperplane with an allowed error margin, ɛ-insensitive tube (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. (b)). Artificial neural networks (ANNs) are designed to mimic the structure of the human brain, allowing it to process and interpret vast amounts of data into information that can be put to use (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(c)). Therefore, ANNs are frequently utilized as they work well for complex and nonlinear systems (Sharifzadeh et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Random Forest (RF) could be used for both classification and regression. RF uses multiple trees to predict the majority of the modes and the average prediction for clarification and regression problems, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(d)). Details of RF are explained well in 2019 by Fan et. al. (Xing et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2019b\u003c/span\u003e). Gradient boosting (GB) can be used for both classification and regression. In boosting, several weak learners are created sequentially to enhance the overall performance of the method (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(e)). GB minimizes a loss function by training an ensemble of predictors one after the other, compensating for the mistakes of the preceding forecasters (Hastie et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2001\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eUnsupervised learning can be used for feature extraction. For example, Aghbashlo et al. collected a sludge pyrolysis dataset and applied principal component analysis to decrease dimensionality (Shahbeik et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Supervised and unsupervised learning find hidden patterns of datasets, but they do not decide action. RL can decide optimal action considering uncertainty. RL allows sequential decision making and is used for various purposes such as control, scheduling, and planning (Lee et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"3. ML applications in bio-oil research","content":"\u003cp\u003eIn the last few years, various ML algorithms have been investigated to predict the performance of pyrolysis. The primary objective of previous research is to predict bio-oil yield, contents, heating values based on contents of biomass, and operating conditions (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The top five ML algorithms used for modelling of bio-oil production are as follows: RF, ANN, GB, SVR, and LR.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe top five ML algorithms for modelling of bio-oil production\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorism\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInput variables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOutput variables\u003c/p\u003e \u003cp\u003e(Prediction targets)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTested method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBest method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePerformance\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eRef.\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Ash\u003c/p\u003e \u003cp\u003e\u0026bull;Volatiles\u003c/p\u003e \u003cp\u003e\u0026bull;Feedstock O/C ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Oasmaa et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2010\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Gas chromatography\u003c/p\u003e \u003cp\u003e\u0026ndash;mass spectrometry\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;HHV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Wanignon Ferdinand et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2012\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;13C NMR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Mass fractions\u003c/p\u003e \u003cp\u003eof C, H, N and O\u003c/p\u003e \u003cp\u003e\u0026bull;HHV\u003c/p\u003e \u003cp\u003e\u0026bull;Phenol \u0026amp; cresols conc.\u003c/p\u003e \u003cp\u003e\u0026bull; Total acid number\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePLS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePLS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Strahan et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2016\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Sludge ultimate\u003c/p\u003e \u003cp\u003ecomposition\u003c/p\u003e \u003cp\u003e\u0026bull;Proximate\u003c/p\u003e \u003cp\u003ecompositions\u003c/p\u003e \u003cp\u003e\u0026bull;Temperature\u003c/p\u003e \u003cp\u003e\u0026bull;Heating rate\u003c/p\u003e \u003cp\u003e\u0026bull;Reaction time\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Bio-oil yield\u003c/p\u003e \u003cp\u003e\u0026bull;Syngas yield\u003c/p\u003e \u003cp\u003e\u0026bull;Syngas composition\u003c/p\u003e \u003cp\u003e\u0026bull;Biochar yield\u003c/p\u003e \u003cp\u003e\u0026bull;Biochar atomic\u003c/p\u003e \u003cp\u003ecomposition\u003c/p\u003e \u003cp\u003e\u0026bull;Biochar calorific value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMLPNN, SVR, RFR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.813\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Shahbeik et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Feedstock\u003c/p\u003e \u003cp\u003ecomposition\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003cp\u003e\u0026bull;Hydrogen content\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMLR, RF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Tang et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2020\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Biomass compositions\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003cp\u003e\u0026bull;Viscosity, HHV\u003c/p\u003e \u003cp\u003e\u0026bull;O/C and H/C ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Zhang et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Temperature\u003c/p\u003e \u003cp\u003e\u0026bull;Heating rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Bio-char yield\u003c/p\u003e \u003cp\u003e\u0026bull;Bio-oil yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN, RSM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.922\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Angın and Tiryaki \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2016\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Feedstock\u003c/p\u003e \u003cp\u003ecomposition\u003c/p\u003e \u003cp\u003e\u0026bull;Catalyst type and ratio\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Bio-oil yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.933\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(\u0026Ouml;ZBAY and K\u0026Ouml;KTEN \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2020\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Temperature\u003c/p\u003e \u003cp\u003e\u0026bull;Heating rate\u003c/p\u003e \u003cp\u003e\u0026bull;Conversion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Activation energy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.999\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Asghar et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN, GNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.9924\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Singh et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSVR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Feedstock\u003c/p\u003e \u003cp\u003ecomposition\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Three-phase\u003c/p\u003e \u003cp\u003eproduct distribution\u003c/p\u003e \u003cp\u003e\u0026bull;HHV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN, SVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Chen et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Feedstock\u003c/p\u003e \u003cp\u003echaracterization\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSVR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Potnuri et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eGB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Biomass compositions\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Bio-oil yield\u003c/p\u003e \u003cp\u003e\u0026bull;Biochar yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGB, DNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eXGB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Alabdrabalnabi et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Biomass compositions\u003c/p\u003e \u003cp\u003e\u0026bull;Microwave power\u003c/p\u003e \u003cp\u003e\u0026bull;Temperature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003cp\u003e\u0026bull;Product contents, HHV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVR, RF, GBR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGBR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.823\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Yang et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Feedstock type\u003c/p\u003e \u003cp\u003e\u0026bull;Feedstock\u003c/p\u003e \u003cp\u003ecomposition\u003c/p\u003e \u003cp\u003e\u0026bull;Temperature\u003c/p\u003e \u003cp\u003e/heating rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Char, oil, gas yield\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN, GBR, DT, RF, KNN, BR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGBR/BR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Shen et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull;Biomass compositions\u003c/p\u003e \u003cp\u003e\u0026bull;Operating conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull;Yield\u003c/p\u003e \u003cp\u003e\u0026bull;Nitrogen heterocycles\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF, GBR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e(Leng et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the early stage, simple LR was used for bio-oil yield prediction. Oasmaa et al. used a multiple linear regression model to explore the correlation between bio-oil yield and feedstock characteristics (ash, volatiles, and O/C ratio) (Oasmaa et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Ferdinand et al. predicted higher heating value (HHV) of bio-oil based on GC\u0026ndash;MS analysis data by using a conventional multiple regression model (Wanignon Ferdinand et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Mullen et al. used the partial least-squares (PLS) regression model to predict the HHV of bio-oil based on the 13C NMR analysis information (Strahan et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Rao et al. used a polynomial regression algorithm to estimate the impact of catalyst quantity and pre-treatment temperature on yield, heating rate, pyrolysis temperature, and conversion efficiency (Potnuri et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eMeanwhile, more advanced method were used for predicting the yield of biochar (Cao et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), gas production selectivity (Sun et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), and bio-diesel production (Mostafaei et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). To the best of our knowledge, in 2016, Angin et al. became the first group to apply the ANN method to predict the yield of bio-oil produced through pyrolysis and they concluded that the ANN model can replace the response surface methodology due to its high accuracy and generalization capability (Angın and Tiryaki \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). ML studies for the prediction of bio-oil production and its characteristics via thermochemical reaction have been actively underway since 2020, as shown in Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eOzbay et al. developed an ANN model to predict the bio-oil yield based on proximate and ultimate analyses of biomass, catalyst type, and operating conditions. The slow and intermediate pyrolysis data were used and the prediction results were quite consistent with experiment results (\u0026Ouml;ZBAY and K\u0026Ouml;KTEN \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Mehmood et al. employed a multilayer perceptron-based ANN regression model for determining the activation energy of Saccharum Bengalense pyrolysis (Asghar et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Singh et al. applied an ANN and generalized neural network (GNN) to predict bio-oil yield from lychee-based biomass based on pyrolysis parameters such as temperature, gas flow, heating rate, and retention time. GNN offered more precision in comparison to ANN (Singh et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSVR was also often utilized to forecast product attributes and yield. Yuan et al. forecasted the biochar yield from cattle manure utilizing a least-squares support vector machine and an ANN based on 33 experimental datasets (Cao et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Xiao et al. developed prediction models for the three-phase product distribution and bio-oil heating value using an ANN and SVM. They reported that SVR showed better performance (Chen et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Rao et al. successfully utilized a SVM to predict the product yields from the co-pyrolysis of biomass and plastics. This study considered the catalyst and blend as input features (Potnuri et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe RF method is the most often used ML algorithm for biomass research. It was mainly applied to predict the composition contents (cellulose, hemicellulose, and lignin) (Xing et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2019b\u003c/span\u003e) and HHV (Xing et al. \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2019a\u003c/span\u003e) of biomass and to forecast the yield of bio-char produced by pyrolysis (Zhu et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Yang et al. used the RF method for predicting bio-oil yield and hydrogen content of bio-oil, based on the types of biomass and pyrolysis reaction conditions. Feedstock compositions had a greater an impact on both yield and hydrogen content (Tang et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In 2021, Yang et al. used the RF method to predict gas yield and its major composition. They reported that pyrolysis conditions contributed more to predict yield and H\u003csub\u003e2\u003c/sub\u003e and CO\u003csub\u003e2\u003c/sub\u003e concentration (Tang et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Wang et al. used the RF method to develop prediction models for predicting the yield and carbon content of biochar based on the pyrolysis data of lignocellulosic biomass. The yield and C-char changes were primarily attributed to the pyrolysis temperature (Zhu et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Mu et al. used RF to predict yield, HHV, viscosity, and oxygen and carbon ratio of bio-oil. They reported that pyrolysis conditions (temperature, heating rate, and particle size) is less influential than the ultimate and proximate analysis of feedstock (Zhang et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In previous studies, the most important input characteristic differed depending on the dataset. The variety of feedstocks and pyrolysis conditions leads to an inconsistent input/output correlation, and this highlights the significance of the dataset.\u003c/p\u003e \u003cp\u003eRecently, GB methods were frequently used. Gautam et al. developed a dense neural network and the XGBoost (XGB) model to predict the yield of co-pyrolysis of biomass and waste plastic. XGB offered the best performance with a root mean squared error of 1.77 and an R\u003csup\u003e2\u003c/sup\u003e of 0.96 (Alabdrabalnabi et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Tabatabaei compared SVR, RF, and gradient boosting regressor (GBR) for microwave-assisted pyrolysis. Among them, GBR provides better prediction performance with a R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.822 (Yang et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Gao et al. used ANN, GB, decision trees (DT), RF, K-nearest-neighbors, bagging regressor, and lasso regression to predict char, bio-oil, and gas product yields. GB provided better predictions with a R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.90 in single output models, whereas the bagging regressor showed the best performance in multiple output models (Shen et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Wu et al. developed the machine-learning assisted Aspen Plus rector model, which predicts the yield, HHV, and yield component distribution of biocrude oil according to the experimental records of the hydrothermal liquefaction process from 16 species of microalgae. The least-square boosting model showed the best performance for the prediction (Wu et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Yang et al. compared RF and GBR in terms of prediction of nitrogen heterocycles of bio-oil. For this, 217 datasets were collected from 63 SCI papers comprising 91 biomasses and the Pearson correlation coefficient was used. In this study, RF returned better prediction than GBR for every case (Leng et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Various ML methods will be used based on the advancement of the ML algorithm. This will be discussed in the next section.\u003c/p\u003e \u003cp\u003eAdditionally, Colosi et al. made further improvement by integrating ML, life cycle assessment (LCA), and techno-economic analysis (TEA). The RF model predicts the yield and characteristics of bio-char, and calculates the parameters of LCA and TEA such as energy return on investment, global warming potential, and minimum product selling price (Cheng et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Chemmangattuvalappil et al. developed a rule-based model using rough set machine learning (RSML), which allows to combine expert knowledge and a dataset. In this study, 207 and 128 data points in published studies were used to estimate the HHV and pH of bio-oil, respectively (Chong et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Xiong et al. used historical computational fluid dynamics (CFD) data to build a yield prediction model. The long and short term memory network was used to predict the mass flow rates at the reactor. The CFD simulation time was reduced by almost 30% through the application of ML (Zhong et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Khan et al. applied five different ML methods (SVM, ANN, DT, Gaussian process regression, and ensembled tree) integrated with two optimization methods (particle swarm optimization and genetic algorithm) for feature selection and hyperparameter optimization of ML models. A Gaussian process regression - genetic algorithm model showed the best performance (Ullah et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Baskar et al investigated role of AI technologies in the municipal solid waste management area. They showed that AI can play a significant role not only for the prediction of reaction parameters but also the sorting process (Naveenkumar et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn summary, previous studies mainly used ML to predict bio-oil and gas yield and the component and characteristics of products based on feedstock type, components, operating conditions, and pyrolysis type. Therefore, most ML applications have been regressions. Frequently, users extract features using their own knowledge. This also means that the datasets that were used were limited. The limitations of previous studies include use of confined data and limited use of ML algorithms for the prediction of the pyrolysis reaction.\u003c/p\u003e"},{"header":"4. Future Outlook","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Dataset and DL\u003c/h2\u003e \u003cp\u003eBio-oil can be produced through various pyrolysis conditions and feedstocks. Numerous bio-oil datasets, produced through both experimentation and modeling/simulation, are obtainable in the literature. However, in previous studies, most datasets learned by ML were collected by users with specific goals in mind. In other words, the confined datasets were used for specific purposes, and this resulted in the usage of constrained algorithms, namely regression. For ML applications, data collection is always essential. Both the quality and quantity of data are important. Learning a sufficient number of qualified datasets by ML may provide new insight, including possibly even counter-intuitive findings. If there are enough labeled datasets, then DL techniques can be used. DL can tackle complex problems even when the datasets are exceedingly diverse and unstructured (Taye \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). It is expected that DL would be advantageous to the bio-oil research community.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Hybrid model\u003c/h2\u003e \u003cp\u003eIn previous studies, researchers utilized their knowledge only for problem formulation (i.e. define input/output variables), and not problem solving. In order to address this deficiency, Chemmangattuvalappil et al. adapted the RSML approach (Chong et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Rule-based model ML approaches enable the merging of expert knowledge with data incorporated in information throughout the training process. Namely, domain knowledge may be added through user-specified training parameters to guarantee that the final rules make sense in terms of physical mechanisms. The black-box ML techniques have low intrinsic interpretability, poor extrapolating capabilities, and unbounded uncertainty in predictions. Attempts to address the deficiency include a hybrid model that learns from data and physics. In other words, combining ML with first-principles knowledge is the next horizon since it can increase accuracy and interpretability while utilizing less data. The hybrid concept was well articulated by Boukouvala (Bradley et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Optimal decision making: process, control, scheduling, design of experiment\u003c/h2\u003e \u003cp\u003eML is capable of making optimal decisions. In particular, RL learns the optimal decision policy and links the system state to the optimal course of action. As decisions made at one time have an impact on subsequent decisions and the resulting output events, this is a natural feature of a dynamical system; this is where RL differs from SL. RL uses evaluative feedback from the environment to estimate real-valued rewards or costs, whereas SL uses informative feedback through classification errors (Lee et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Therefore, ML may be applied to optimal control, planning, and scheduling.\u003c/p\u003e \u003cp\u003eAdditionally, in terms of design of experiment, Bayesian optimization (BO) is a useful tool. It has been shown that BO is sample-efficient and scalable, requiring minimal testing and enabling the exploration of large design spaces (Gonz\u0026aacute;lez and Zavala \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). BO can be employed in bio-oil pyrolysis experiments to reduce trial and error.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eA comprehensive review of bio-oil pyrolysis research using ML has been performed. In the last decade, numerous ML approaches have been studied to forecast pyrolysis performance. Target variables were mainly bio-oil and gas yield, contents, and chemical properties of products, and predictors were feedstock type, ultimate and proximate composition of feedstock, reactor type, and operating conditions such as temperature and retention time. The most frequently used ML was regression algorithms: RF, ANN, GB, SVR, and LR. The limitations of previous studies are use of confined data and limited use of ML algorithms for prediction of the pyrolysis reaction. We anticipate that ML will aid the bio-oil research field in further utilizing DL, hybrid models, RL, and BO.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eML \u0026nbsp;Machine Learning\u003c/p\u003e\n\u003cp\u003eAI \u0026nbsp;Artificial Intelligence\u003c/p\u003e\n\u003cp\u003eDL \u0026nbsp;Deep Learning\u003c/p\u003e\n\u003cp\u003eSL \u0026nbsp;Supervised Learning\u003c/p\u003e\n\u003cp\u003eRL \u0026nbsp;Reinforcement Learning\u003c/p\u003e\n\u003cp\u003eLR \u0026nbsp;Linear Regression\u003c/p\u003e\n\u003cp\u003eSVM \u0026nbsp;Support Vector Machine\u003c/p\u003e\n\u003cp\u003eSVR \u0026nbsp;Support Vector Regression\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eANNs \u0026nbsp;Artificial Neural Networks\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRF \u0026nbsp;Random Forest\u003c/p\u003e\n\u003cp\u003eGB \u0026nbsp;Gradient Boosting\u003c/p\u003e\n\u003cp\u003eHHV \u0026nbsp;Higher Heating Value\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePLS \u0026nbsp;Partial Least-Squares\u003c/p\u003e\n\u003cp\u003eGNN \u0026nbsp;Generalized Neural Network\u003c/p\u003e\n\u003cp\u003eXGB \u0026nbsp;XGBoost\u003c/p\u003e\n\u003cp\u003eGBR \u0026nbsp;Gradient Boosting Regressor\u003c/p\u003e\n\u003cp\u003eDT \u0026nbsp;Decision Tree\u003c/p\u003e\n\u003cp\u003eLCA \u0026nbsp;Life Cycle Assessment\u003c/p\u003e\n\u003cp\u003eTEA \u0026nbsp;Techno-Economic Analysis\u003c/p\u003e\n\u003cp\u003eRSML \u0026nbsp;Rough Set Machine Learning\u003c/p\u003e\n\u003cp\u003eCFD \u0026nbsp;Computational Fluid Dynamics\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBO \u0026nbsp;Bayesian Optimization\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCRediT authorship contribution statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHyojin Lee\u003c/strong\u003e: Conceptualization, Methodology, Investigation, Visualization, Writing – original draft. \u003cstrong\u003eIL-Ho Choi\u003c/strong\u003e: Visualization, Writing – review \u0026amp; editing. \u003cstrong\u003eKyung-Ran Hwang\u003c/strong\u003e: Conceptualization, Supervision, Project administration, Writing – review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of competing interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could appear to influence the work reported in this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo data were used for the research described in the article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No.NRF2020M1A2A2079802).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAppendix A. Supplementary data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupplementary data for this work can be found in the e-version of this paper online.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAbdullah N, Mohd Taib R, Mohamad Aziz NS, et al (2023) Banana pseudo-stem biochar derived from slow and fast pyrolysis process. Heliyon 9:e12940. https://doi.org/10.1016/j.heliyon.2023.e12940\u003c/li\u003e\n\u003cli\u003eAlabdrabalnabi A, Gautam R, Mani Sarathy S (2022) Machine learning to predict biochar and bio-oil yields from co-pyrolysis of biomass and plastics. Fuel 328:125303. https://doi.org/10.1016/j.fuel.2022.125303\u003c/li\u003e\n\u003cli\u003eAngın D, Tiryaki AE (2016) Application of response surface methodology and artificial neural network on pyrolysis of safflower seed press cake. Energy Sources, Part A: Recovery, Utilization, and Environmental Effects 38:1055\u0026ndash;1061. https://doi.org/10.1080/15567036.2013.862585\u003c/li\u003e\n\u003cli\u003eAsghar A, Liu C-G, Ali I, et al (2023) Bioenergy potential of Saccharum bengalense through pyrolysis, reaction kinetics, TG-FTIR-GCMS analysis of pyrolysis products, and validation of the pyrolysis data through machine learning. Chemical Engineering Journal 465:142930. https://doi.org/10.1016/j.cej.2023.142930\u003c/li\u003e\n\u003cli\u003eBradley W, Kim J, Kilwein Z, et al (2022) Perspectives on the integration between first-principles and data-driven modeling. Comput Chem Eng 166:107898. https://doi.org/10.1016/j.compchemeng.2022.107898\u003c/li\u003e\n\u003cli\u003eCao H, Xin Y, Yuan Q (2016) Prediction of biochar yield from cattle manure pyrolysis via least squares support vector machine intelligent approach. Bioresour Technol 202:158\u0026ndash;164. https://doi.org/10.1016/j.biortech.2015.12.024\u003c/li\u003e\n\u003cli\u003eCheng F, Luo H, Colosi LM (2020) Slow pyrolysis as a platform for negative emissions technology: An integration of machine learning models, life cycle assessment, and economic analysis. Energy Convers Manag 223:113258. https://doi.org/10.1016/j.enconman.2020.113258\u003c/li\u003e\n\u003cli\u003eChen X, Zhang H, Song Y, Xiao R (2018) Prediction of product distribution and bio-oil heating value of biomass fast pyrolysis. Chemical Engineering and Processing - Process Intensification 130:36\u0026ndash;42. https://doi.org/10.1016/j.cep.2018.05.018\u003c/li\u003e\n\u003cli\u003eChong JW, Thangalazhy‐Gopakumar S, Tan RR, et al (2022) Estimation of fast pyrolysis bio‐oil properties from feedstock characteristics using rough‐set ‐based machine learning. Int J Energy Res 46:19159\u0026ndash;19176. https://doi.org/10.1002/er.8201\u003c/li\u003e\n\u003cli\u003eGholizadeh M, Zhang S, Hu X, Wang Y (2023) Advances and Perspectives of Bio-oil Hydrotreatment for Biofuel Production. Energy \u0026amp; Fuels 37:10134\u0026ndash;10154. https://doi.org/10.1021/acs.energyfuels.3c01647\u003c/li\u003e\n\u003cli\u003eGonz\u0026aacute;lez LD, Zavala VM (2023) New paradigms for exploiting parallel experiments in Bayesian optimization. Comput Chem Eng 170:108110. https://doi.org/10.1016/j.compchemeng.2022.108110\u003c/li\u003e\n\u003cli\u003eGuedes RE, Luna AS, Torres AR (2018) Operating parameters for bio-oil production in biomass pyrolysis: A review. J Anal Appl Pyrolysis 129:134\u0026ndash;149. https://doi.org/10.1016/j.jaap.2017.11.019\u003c/li\u003e\n\u003cli\u003eHafeez S, Pallari E, Manos G, Constantinou A (2019) Catalytic Conversion and Chemical Recovery. In: Plastics to Energy. Elsevier, pp 147\u0026ndash;172\u003c/li\u003e\n\u003cli\u003eHastie T, Friedman J, Tibshirani R (2001) The Elements of Statistical Learning. Springer New York, New York, NY\u003c/li\u003e\n\u003cli\u003eJahanshahi A, Lopes M, Brand\u0026atilde;o M, De Castro EA (2023) Development of bioenergy technologies: A scientometric analysis. Heliyon 9:e20000. https://doi.org/10.1016/j.heliyon.2023.e20000\u003c/li\u003e\n\u003cli\u003eJanga JK, Reddy KR, Raviteja KVNS (2023) Integrating artificial intelligence, machine learning, and deep learning approaches into remediation of contaminated sites: A review. Chemosphere 345:140476. https://doi.org/10.1016/j.chemosphere.2023.140476\u003c/li\u003e\n\u003cli\u003eLee JH, Shin J, Realff MJ (2018) Machine learning: Overview of the recent progresses and implications for the process systems engineering field. Comput Chem Eng 114:111\u0026ndash;121. https://doi.org/10.1016/j.compchemeng.2017.10.008\u003c/li\u003e\n\u003cli\u003eLehto J, Oasmaa A, Solantausta Y, et al (2014) Review of fuel oil quality and combustion of fast pyrolysis bio-oils from lignocellulosic biomass. Appl Energy 116:178\u0026ndash;190. https://doi.org/10.1016/j.apenergy.2013.11.040\u003c/li\u003e\n\u003cli\u003eLeng L, Li T, Zhan H, et al (2023) Machine learning-aided prediction of nitrogen heterocycles in bio-oil from the pyrolysis of biomass. Energy 278:127967. https://doi.org/10.1016/j.energy.2023.127967\u003c/li\u003e\n\u003cli\u003eMostafaei M, Javadikia H, Naderloo L (2016) Modeling the effects of ultrasound power and reactor dimension on the biodiesel production yield: Comparison of prediction abilities between response surface methodology (RSM) and adaptive neuro-fuzzy inference system (ANFIS). Energy 115:626\u0026ndash;636. https://doi.org/10.1016/j.energy.2016.09.028\u003c/li\u003e\n\u003cli\u003eNaveenkumar R, Iyyappan J, Pravin R, et al (2023) A strategic review on sustainable approaches in municipal solid waste management and energy recovery: Role of artificial intelligence, economic stability and life cycle assessment. Bioresour Technol 379:129044. https://doi.org/10.1016/j.biortech.2023.129044\u003c/li\u003e\n\u003cli\u003eOasmaa A, Solantausta Y, Arpiainen V, et al (2010) Fast Pyrolysis Bio-Oils from Wood and Agricultural Residues. Energy \u0026amp; Fuels 24:1380\u0026ndash;1388. https://doi.org/10.1021/ef901107f\u003c/li\u003e\n\u003cli\u003e\u0026Ouml;ZBAY G, K\u0026Ouml;KTEN ES (2020) Modeling of Bio-Oil Production by Pyrolysis of Woody Biomass: Artificial Neural Network Approach. Politeknik Dergisi 23:1255\u0026ndash;1264. https://doi.org/10.2339/politeknik.659136\u003c/li\u003e\n\u003cli\u003ePotnuri R, Rao CS, Surya DV, et al (2023) Utilizing support vector regression modeling to predict pyro product yields from microwave-assisted catalytic co-pyrolysis of biomass and waste plastics. Energy Convers Manag 292:117387. https://doi.org/10.1016/j.enconman.2023.117387\u003c/li\u003e\n\u003cli\u003ePotnuri R, Suriapparao D V., Sankar Rao C, et al (2022) Effect of dry torrefaction pretreatment of the microwave-assisted catalytic pyrolysis of biomass using the machine learning approach. Renew Energy 197:798\u0026ndash;809. https://doi.org/10.1016/j.renene.2022.08.006\u003c/li\u003e\n\u003cli\u003eShahbeik H, Rafiee S, Shafizadeh A, et al (2022) Characterizing sludge pyrolysis by machine learning: Towards sustainable bioenergy production from wastes. Renew Energy 199:1078\u0026ndash;1092. https://doi.org/10.1016/j.renene.2022.09.022\u003c/li\u003e\n\u003cli\u003eSharifzadeh M, Sikinioti-Lock A, Shah N (2019) Machine-learning methods for integrated renewable power generation: A comparative study of artificial neural networks, support vector regression, and Gaussian Process Regression. Renewable and Sustainable Energy Reviews 108:513\u0026ndash;538. https://doi.org/10.1016/j.rser.2019.03.040\u003c/li\u003e\n\u003cli\u003eShen J, Yan M, Fang M, Gao X (2022) Machine learning-based modeling approaches for estimating pyrolysis products of varied biomass and operating conditions. Bioresour Technol Rep 20:101285. https://doi.org/10.1016/j.biteb.2022.101285\u003c/li\u003e\n\u003cli\u003eSingh Y, Singh D, Singh NK, et al (2023) Production of bio-oil from lychee-based biomass through pyrolysis and maximization of bio-oil yield with statistical and machine learning techniques. J Clean Prod 413:137472. https://doi.org/10.1016/j.jclepro.2023.137472\u003c/li\u003e\n\u003cli\u003eStrahan GD, Mullen CA, Boateng AA (2016) Prediction of Properties and Elemental Composition of Biomass Pyrolysis Oils by NMR and Partial Least Squares Analysis. Energy \u0026amp; Fuels 30:423\u0026ndash;433. https://doi.org/10.1021/acs.energyfuels.5b02345\u003c/li\u003e\n\u003cli\u003eSun Y, Liu L, Wang Q, et al (2016) Pyrolysis products from industrial waste biomass based on a neural network model. J Anal Appl Pyrolysis 120:94\u0026ndash;102. https://doi.org/10.1016/j.jaap.2016.04.013\u003c/li\u003e\n\u003cli\u003eTang Q, Chen Y, Yang H, et al (2020) Prediction of Bio-oil Yield and Hydrogen Contents Based on Machine Learning Method: Effect of Biomass Compositions and Pyrolysis Conditions. Energy \u0026amp; Fuels 34:11050\u0026ndash;11060. https://doi.org/10.1021/acs.energyfuels.0c01893\u003c/li\u003e\n\u003cli\u003eTang Q, Chen Y, Yang H, et al (2021) Machine learning prediction of pyrolytic gas yield and compositions with feature reduction methods: Effects of pyrolysis conditions and biomass characteristics. Bioresour Technol 339:125581. https://doi.org/10.1016/j.biortech.2021.125581\u003c/li\u003e\n\u003cli\u003eTaye MM (2023) Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Future Directions. Computers 12:91. https://doi.org/10.3390/computers12050091\u003c/li\u003e\n\u003cli\u003eUllah H, Haq ZU, Naqvi SR, et al (2023) Optimization based comparative study of machine learning methods for the prediction of bio-oil produced from microalgae via pyrolysis. J Anal Appl Pyrolysis 170:105879. https://doi.org/10.1016/j.jaap.2023.105879\u003c/li\u003e\n\u003cli\u003eWanignon Ferdinand F, Van de Steene L, Kamenan Blaise K, Siaka T (2012) Prediction of pyrolysis oils higher heating value with gas chromatography\u0026ndash;mass spectrometry. Fuel 96:141\u0026ndash;145. https://doi.org/10.1016/j.fuel.2012.01.007\u003c/li\u003e\n\u003cli\u003eWu W, Huang C-M, Tsai Y-H (2023) Design and validation of a microalgae biorefinery using machine learning-assisted modeling of hydrothermal liquefaction. Algal Res 74:103230. https://doi.org/10.1016/j.algal.2023.103230\u003c/li\u003e\n\u003cli\u003eXing J, Luo K, Wang H, et al (2019a) A comprehensive study on estimating higher heating value of biomass from proximate and ultimate analysis with machine learning approaches. Energy 188:116077. https://doi.org/10.1016/j.energy.2019.116077\u003c/li\u003e\n\u003cli\u003eXing J, Luo K, Wang H, Fan J (2019b) Estimating biomass major chemical constituents from ultimate analysis using a random forest model. Bioresour Technol 288:121541. https://doi.org/10.1016/j.biortech.2019.121541\u003c/li\u003e\n\u003cli\u003eYang Y, Shahbeik H, Shafizadeh A, et al (2022) Biomass microwave pyrolysis characterization by machine learning for sustainable rural biorefineries. Renew Energy 201:70\u0026ndash;86. https://doi.org/10.1016/j.renene.2022.11.028\u003c/li\u003e\n\u003cli\u003eZhang T, Cao D, Feng X, et al (2022) Machine learning prediction of bio-oil characteristics quantitatively relating to biomass compositions and pyrolysis conditions. Fuel 312:122812. https://doi.org/10.1016/j.fuel.2021.122812\u003c/li\u003e\n\u003cli\u003eZhang Z, Matharu AS (2018) Thermochemical Valorization of Paper Deinking Residue through Microwave-Assisted Pyrolysis. In: Waste Biorefinery. Elsevier, pp 671\u0026ndash;692\u003c/li\u003e\n\u003cli\u003eZheng J-L, Zhu Y-H, Zhu M-Q, et al (2019) A review of gasification of bio-oil for gas production. Sustain Energy Fuels 3:1600\u0026ndash;1622. https://doi.org/10.1039/C8SE00553B\u003c/li\u003e\n\u003cli\u003eZhong H, Wei Z, Man Y, et al (2023) Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data. J Clean Prod 391:136192. https://doi.org/10.1016/j.jclepro.2023.136192\u003c/li\u003e\n\u003cli\u003eZhu X, Li Y, Wang X (2019) Machine learning prediction of biochar yield and carbon contents in biochar based on biomass characteristics and pyrolysis conditions. Bioresour Technol 288:121527. https://doi.org/10.1016/j.biortech.2019.121527\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"machine learning, predictive modeling, pyrolysis, bio-oil, lignocellulose","lastPublishedDoi":"10.21203/rs.3.rs-3830648/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3830648/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBio-oil produced through pyrolysis of lignocellulosic biomass has recently received significant attention due to its possible uses as a second-generation biofuel. The yield and characteristics of produced bio-oil are affected by reaction conditions (reactor type, particle size, feed rate, operating temperature, heating rate, retention time, etc.) and the type of feedstock that is used (softwood, hardwood, agricultural plant residues, miscanthus, etc.). Recently, machine learning (ML) techniques have been widely employed to forecast the performance of the pyrolysis and the characteristics of bi-oil. In this study, a comprehensive review of ML research on bio-oil has been carried out. Regression methods were most frequently employed to build prediction models. The top five ML methods for bio-oil research were random forest, artificial neural network, gradient boosting, support vector regression, and linear regression. In addition, users frequently extract features using their own knowledge and restricted datasets were employed I previous studies. We highlighted the challenges and potential of cutting-edge ML techniques in bio-oil production.\u003c/p\u003e","manuscriptTitle":"A comprehensive review of machine learning prediction in the production of bio-oil from lignocellulose via pyrolysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-04 10:03:15","doi":"10.21203/rs.3.rs-3830648/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":"05997117-7ebe-4c0a-876f-28d8bfd7d136","owner":[],"postedDate":"January 4th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-01-12T02:59:19+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-04 10:03:15","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3830648","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3830648","identity":"rs-3830648","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}