Optimisation and performance evaluation of response surface methodology (RSM), artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) in the prediction of biogas production from palm oil mill effluent (POME)

In recent years, machine learning (ML) techniques have been developed to predict the performance of anaerobic digestion (AD) processes including methane potential and reactor stability. However, their practical applications to industrial-scale palm oil mill effluent (POME) treatment plant are limited. In this study, ML algorithms such as response surface methodology (RSM), adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN) are employed to model the biogas production and methane yield from the AD of POME in a local industry-scale anaerobic covered lagoon. Results demonstrated that these models were well aligned with two years of operational data with high coefficient of determination (R2) of up to 0.98. ANFIS yields the highest prediction accuracy, with R2 of 0.9791 along with the lowest mean absolute error (MAE) of 0.0730 and root mean squared error (RMSE) of 0.1438. Subsequently, ANFIS is used in the multi-objective optimisation to maximise the biogas production and methane yield. Optimal conditions for the temperature of the anaerobic digester, pH and recirculation ratio are 38.9 degrees C, 7.03 and 1.89 respectively which could enhance the biogas production and methane yield by 19.4% and 12.2% respectively. Confirmatory experiments were carried out in the biogas plant under this set of optimised variables for a period of two months. The predicted biogas production and methane yield are highly correlated to the actual data with small percentage difference of 1.25% and 5.09% respectively, indicating that ANFIS model was accurate and reliable. Sensitivity analysis shows that pH has the most dominant effect on the methane yield.