Simulation of anaerobic digestion process under variable feeding sludge using a hybrid machine learning model

Anaerobic digestion (AD) is a key method for stabilizing sludge in wastewater treatment plants. With advancements in artificial intelligence, this study investigates a hybrid approach to simulate the performance of AD systems using data from a laboratory-scale setup. Five machine learning (ML) techniques-decision tree (DT), random forest (RF), gradient boosting (GB), support vector regression (SVR), and multilayer perceptron (MLP)-were employed to estimate biogas production, the volatile solids removal, and the total solids output of the digester based on varying input sludge conditions. Model performance was evaluated using the coefficient of determination (R²), mean squared error (MSE), and mean absolute relative error (MARE). The results showed strong correlations between input data and output parameters, with the best model for biogas output achieving an R² of 0.90, MSE of 0.001, and MARE of 5.83 %. In contrast, the lowest performance was observed for total solids output, with an R² of 0.53, MSE of 0.0005, and MARE of 31.46 %. To enhance prediction accuracy, Bayesian model averaging was applied to combine with the results of ML models. This hybrid approach significantly improved the evaluation metrics, increasing the R² from 0.53 to 0.98 and reducing the MSE and MARE to 0.0000 and 4.73 %, respectively.