Characterization of co-digestion of industrial sludges for biogas production by artificial neural network and statistical regression models

The characteristics and impact of industrial sludges of paper, chemical, petrochemical, automobile, and food industries situated in the Ulsan Industrial Complex, Ulsan, Republic of Korea in co-digestion for biogas production were assessed by artificial neural network (ANN) and statistical regression models. The regression model was based on a simplex-centroid mixture design and the ANN was based on a resilient back-propagation algorithm (topology 5-7-1). Using connection weights and bias of the trained ANN model, the impact of each sludge of co-digestion was assessed using Garsons' algorithm. Results suggested that the modelling and predictability of ANN were superior to the regression model with accuracy (A(f)) 1.01, bias (B-f) 1.00, root mean square error 3.56, and standard error of prediction 2.51%. Sludge from the chemical industry showed the highest impact on specific methane yield (SMYvs) with a relative importance of 28.59% followed by sludges from paper (20.07%), food (19.59%), petrochemical (15.92%), and automobile (15.82%) industries. The interactions between diverse industrial sludges were successfully modelled and partitioned into various synergistic and antagonistic effects on SMYvs. Synergistic interactions between the chemical industry sludge and either petrochemical or food industry sludges on SMYvs were detected. However, strong negative interaction between automobile sludge and other sludges was observed. This study indicates that though the ANN model performed better in prediction and impact assessments, the regression model reveals the synergistic and antagonistic interactions among sludges.