Predicting Biogas Yield after Microwave Pretreatment Using Artificial Neural Network Models: Performance Evaluation and Method Comparison

Abstract

In the field of anaerobic digestion (AD) for biogas production, accurately predicting biogas yields following microwave pretreatment (MP) remains a significant challenge. Traditional kinetic models, such as the modified Gompertz (MG) model, are widely utilized but often lack the precision and adaptability needed for optimal process design and operational efficiency. This highlights a crucial gap in the development of more accurate and flexible predictive tools. To address this gap, advanced machine learning techniques, specifically, artificial neural networks (ANN), have been explored. This study developed and evaluated three ANN models: ANN, deep feed forward backpropagation (DFFBP), and deep cascade forward backpropagation network (DCFBP). The DCFBP model demonstrated superior predictive accuracy with a high coefficient of determination (R² = 0.9946) and a lower mean absolute error (MAE = 0.34). Key input parameters, including the ratios of volatile solids to total solids (VS/TS) and the ratio of soluble chemical oxygen demand to total chemical oxygen demand (SCOD/TCOD), were integrated to enhance the prediction precision. These findings highlight the potential of ANN models to improve biogas yield predictions, offering significant benefits for the optimization and design of AD processes.