Prediction and simulation analysis of anaerobic methane production using machine learning methods based on real operational

In this study, data-driven modeling techniques have been developed to estimate anaerobic methane production using real operational data obtained from a functioning biogas plant. The methodological framework is constructed around a sequential prediction chain. In the initial stage of this chain, the reactor's internal characteristics—namely Total Solids (TS), Volatile Solids (VS), Volatile Fatty Acids (VFA), and Total Alkalinity Capacity (TAC)—are estimated based on the input waste quantity and composition. Subsequently, these predicted parameters, in conjunction with waste load, are employed to forecast methane yield. The predictive models are built using a range of Machine Learning (ML) algorithms including Decision Trees (DT), Random Forests (RF), Support Vector Machines (SVM), Gradient Boosting Machines (GBM), and K-Nearest Neighbors (KNN), as well as Artificial Neural Networks (ANN) trained with Levenberg–Marquardt (trainLM), Bayesian Regularization (trainBR), and Scaled Conjugate Gradient (trainSCG) algorithms. Tailored hyperparameter optimization procedures were applied to each model individually. The outcomes indicate that ML-based models generally outperform ANN approaches in terms of accuracy and generalization capacity. Among these, GBM and SVM-based regressors demonstrated superior predictive performance by effectively capturing the nonlinear and complex behavior of the anaerobic digestion system. Additionally, these data-oriented models were benchmarked against classical kinetic models constructed in the MATLAB/Simulink environment. The comparative analysis revealed that fixed-parameter kinetic equations fall short in adequately representing the dynamic behavior of real-world biogas systems. Moreover, evaluation using test sets composed of the monthly minimum and maximum waste inputs confirmed the robustness of the developed models under operational conditions. Accordingly, this work offers an original and comprehensive contribution to the literature by demonstrating the practical applicability and high accuracy potential of machine learning–driven prediction chains in biogas production systems.