Optimizing anaerobic co-digestion of Xyris capensis and duck waste using neuro-fuzzy model and response surface methodology

Optimization of biomethane yield from the anaerobic co-digestion process has taken a new dimension from the application of common statistical techniques to the use of intelligence models in the Artificial Intelligence (AI) field. In this study, the performance of fuzzy c-means (FCM)-clustered adaptive neuro-fuzzy inference system (ANFIS) was compared with the Response Surface Methodology (RSM) in investigating the impact of critical process parameters, namely temperature, retention time, and mixing ratio on the methane yield of anaerobic codigestion of duck waste and Xyris capensis. Relevant statistical metrics such as root mean square error (RMSE), mean absolute percent error (MAPE), mean absolute error (MAE) and mean absolute deviation (MAD) was used to evaluate the performance of the FCM-ANFIS model developed. The developed RSM and FCM-ANFIS models were compared based on the predicted yields and optimum conditions. It was observed from the results that the process parameters have a significant influence on the methane yield of anaerobic co-digestion of duck waste and Xyris capensis. The cumulative yield of 478.42, 478.43, and 436.20 mL CH4/g VSadded was observed for experimental, RSM, and ANFIS-FCM5clusters models, respectively. The optimum daily methane yield was observed when the mixing ratio was 75% duck waste and 25% Xyris capensis for the experimental and ANFIS-FCM5 clusters model. In contrast, the RSM model observed it at 100% duck waste. The RMSE, MAD, MAE, and MAPE values of the optimal neuro-fuzzy model (ANFIS-FCM5clusters) test indicated that the neuro-fuzzy model developed performs better than the RSM model with minimal error and higher accuracy. RSM and ANFIS models were noticed to be practical models for methane yield optimization and prediction from anaerobic co-digestion of livestock waste and lignocellulose feedstocks in a multiple-input variable without undertaking an experiment within a short time and error. These models can be used to optimize the anaerobic digestion process of lignocellulose materials and livestock waste at the industrial scale.