Process modelling and optimisation of methane yield from palm oil mill effluent using response surface methodology and artificial neural network

Abstract

Methane yield plays an important role in evaluating the performance of anaerobic digestion (AD) process due to fluctuating characteristics of palm oil mill effluent (POME). In this study, two years of industrial-scale plant data collected from biogas plant in Pahang, Malaysia consisting of methane yield and H2S concentration were modelled using supervised Machine Learning (ML) algorithms, i.e. artificial neural network (ANN) and response surface methodology (RSM). Input parameters appraised were pH, temperature, and recirculation ratio (RR) of total treated effluent and bottom sludge to raw POME. Fitness of all models was evaluated based on performance metrics, namely determination coefficient (R2) and root mean square error (RMSE). Results demonstrated that ANN was superior to RSM model with R2 of 0.9762 and 0.8500, respectively. Besides, three-dimensional (3-D) response surface plot was utilised to study interactive effects between AD parameters on methane yield. Temperature and recirculation ratio (RR) was found to be main interactive factor of methane yield which was corroborated using Pareto chart. Process optimisation using RSM revealed that the highest methane yield was 0.2733 N m3 CH4/kg CODremoved (i.e. improvement percentage of 9.3%) with the least H2S concentration of 1086.4 mg/L under optimum operating conditions, i.e. temperature, pH and RR of 42 °C, 7.36 and 2.5, respectively. Sensitivity analysis revealed that RR plays substantial roles in yielding more methane in AD process. Maintaining RR ratio at optimum level (1.05 to 2.60) is key to achieve high methane yield with good stability by considering the trade-off between operating cost and revenue of biogas plant.