Enhancing Biogas Yield Prediction Using Artificial Neural Networks: A Case Study with Agricultural Residues

Abstract

Accurate prediction of biogas yield is fundamentally important for the proper running of anaerobic digestion (AD) processes, and it involves complex interactions between process parameters and very nonlinear interactions. To address this problem, the present study systematically developed and evaluated different ANN models for predicting biogas yield from the anaerobic digestion of agricultural residues, including rice straw, cattle manure (CM), and palm oil residues. In all, 200 experimental observations were used to develop the models, with feed pH, temperature, volatile solids (%TS), C/N ratio, and HRT (days) as input variables and biogas yield as the output variable. Five ANN models with different network depths and training algorithms were developed and evaluated. The various performance criteria used for model validation were R² (coefficient of determination), RMSE (root mean squared error), and MAPE (mean absolute percentage error). The experimental results showed that deep ANN models outperformed shallow architectures in modeling the nonlinear trends of the AD phenomenon. Among the tested models, the multi-layered ANN-4 with the Levenberg-Marquardt algorithm achieved the highest predictive accuracy (R² = 0.962, RMSE = 0.018 m³/kg VS, MAPE = 3.1%). Regression analysis and 10-fold cross-validation verified the generalization ability of ANN-4. Sensitivity analysis showed that the C/N ratio, volatile solids (%TS) and HRT were the most significant parameters affecting biogas yield, consistent with established anaerobic digestion mechanisms. Overall, the findings indicate that deep ANN models can be effective tools for optimizing agricultural residue-based anaerobic digestion systems.