Modeling the performance of upflow anaerobic filter (UAF) reactors treating paper-mill wastewater using neural networks

This study investigated the predictive ability of neural networks in the estimation of methane yield (MY) and effluent substrate (Se as mg/L COD) produced by two anaerobic filters, one mesophilic (35°C) and one thermophilic (55°C), which were operated to treat paper-mill wastewater at varying organic loadings. An artificial neural network (ANN) architecture was optimized to obtain a three-layer neural network, composed of three inputs, namely hydraulic retention time (HRT), organic loading rate (OLR), and influent substrate (Si as mg/L COD), six hidden neurons and one output neuron, Se or MY. StoverKincannon model and Multi-linear regression (MLR) technique was also used for data analysis and to compare the prediction capability. Four statistical criteria also used for comparison were mean square error (MSE), mean absolute error (MAE), mean absolute relative error (MARE), and determination coefficient (R 2 ). The results showed that ANN approach predicted the performance of the anaerobic filters better than both Stover-Kincannon model and MLR technique.