Forecasting Of Biogas And Biomethane Outputs From Anaerobic Co-Digestion Using Multilayer Perceptron Artificial Neural Networks (Mlp-Ann)

The intricate, nonlinear interactions between several process parameters make it difficult to forecast biogas and biomethane (bioCH₄) yields in anaerobic digestion systems with any degree of accuracy. These dynamics are frequently not well captured by conventional kinetic models, particularly in co-digestion systems with heterogeneous substrates. This was addressed by developing and testing a Multilayer Perceptron Artificial Neural Network (MLP-ANN) for predicting cumulative biogas and bioCH₄ volumes from the anaerobic co-digestion of soymilk dregs and cow manure based on key operational parameters: pH, Substrate Mixing Ratio (SMR), and Hydraulic Retention Time (HRT) while the temperature was constant (33 ± 1 °C). An MLP architecture with a 3-3-2 network structure was employed, comprising three input neurons, two hidden layers, and two output neurons. The model was trained and validated using standardized input variables, with hyperbolic tangent activation in the hidden layers and identity activation in the output layer. The ANN model demonstrated excellent predictive accuracy, achieving average R² values of 0.980 for biogas and 0.976 for BioCH₄ during five-fold cross-validation, with correspondingly low mean absolute error (MAE) and root mean square error (RMSE) values. Parameter estimates indicated that HRT had the most significant influence on gas production, while SMR and pH served as important supporting factors. Practical recommendations include optimizing HRT and SMR settings and maintaining stable pH levels to maximize biogas system efficiency. The study highlights the importance of ANN models for improving operational planning and optimizing biogas production processes. The developed model offers a robust tool for predicting energy recovery outcomes in anaerobic digestion systems and supports the broader adoption of machine learning in biogas process optimization.