Artificial Neural Network Predictions of Up-Flow Anaerobic Sludge Blanket (UASB) Reactor Performance in the Treatment of Citrus Juice Wastewater

Abstract

The operation of a full-scale up-flow anaerobic sludge blanket (UASB) reactor treating citrus juice wastewater was observed for two years. The average total chemical oxygen demand (COD) removal efficiency was determined to be equal to 79% and 77%, for the first and second years of operation for this reactor, respectively. The average volumetric loading rate was equal to 8.1 and 5.7 kg COD/m(3)day, respectively, during these periods. Three artificial neural network (ANN) models, namely feed forward back propagation (FFBP), radial basis function-based neural networks (RBF), and generalized regression neural networks (GRNN) were utilized to predict the COD and total suspended solid (TSS) concentrations in the effluent leaving the UASB reactor as well as the biogas production in the reactor. In general, the FFBP model made the best predictions with an average deviation of about 6.4-15.6% from the experimental values. The predictions made for biogas production and COD concentration were more accurate, while relatively larger discrepancies existed for the TSS concentration. The utilization of the ANN models generally provided significant improvements when compared to the use of multilinear regression for the same purpose.