Optimisation of anaerobic digestion of palm oil mill effluent with biochar addition: Synergistic application of Artificial neural network and response Surface Methodology

• 104 % increase in methane yield from anaerobic co-digestion of POME with biochar. • Integration of RSM and ANN for enhanced process optimization. • RSM identified the non-linear relationship between BC dosage and methane yield. Fluctuations in biogas production and low methane yields remain significant challenges in the palm oil industry. To address these challenges, this study introduces a novel approach by applying biochar (BC) assisted anaerobic digestion (AD) to palm oil mill effluent (POME) to optimize methane yield. Artificial Neural Network (ANN) modelling and Response Surface Methodology (RSM) were applied to analyze and optimize key process parameters: feed-to-inoculum (FI) ratio, BC dosage, and organic loading (OL). The ANN model demonstrated high predictive accuracy for cumulative biogas volume, achieving an R 2 of 0.9999, MAE of 1.5 mL, MSE of 143 mL 2 , and RMSE of 11.96 mL. For methane yield, ANN recorded an MAE of 0.9 L CH 4 /g VS, MSE of 2.5 L 2 CH 4 /g 2 VS, and RMSE of 1.58 L CH 4 /g VS. Although RSM provided valuable insights into parameter interactions, its predictive accuracy was lower, with R 2 = 0.8917 for biogas volume and R 2 = 0.7534 for methane yield, alongside higher error measures for methane yield (MSE = 95 L 2 CH 4 /g 2 VS). While RSM achieved a lower MSE (45 mL 2 ) for biogas volume, ANN’s higher R 2 and lower MAE suggest better overall predictive capability. Process optimization via RSM identified an optimal FI ratio (0.615), BC dosage (5.36 g/L), and OL (4.8 g VS/L). This led to a 104 % improvement in methane yield from AD of POME with biochar while minimizing BC usage to enhance both biogas production and economic feasibility. These findings highlight the complementary strengths of ANN and RSM, where ANN enables accurate prediction, while RSM facilitates parameter optimization with cost considerations. The results also demonstrate ANN’s potential for real-time process monitoring in industrial AD systems. Future studies should explore alternative BC feedstocks and production conditions to further enhance AD scalability, efficiency, and economic viability.