Estimation of effluent parameters of slaughterhouse wastewater treatment with artificial neural network and B-spline quasi interpolation

Abstract

Effluent of slaughterhouse wastewater treatment by combined up-flow anaerobic sludge blanket (UASB) reactor and extended aeration reactor was estimated through artificial neural networks (ANN), ANN-genetic algorithm (GA) and B-spline quasi interpolation. The overall system operated at two runs with average total chemical oxygen demand (TCOD) of 1514.65 and 3160.2 mg/L for the first and second run, respectively; with two overall hydraulic retention times of 73 and 104 h for each run. The overall system could remove TCOD, ammonia, phosphate and turbidity to a high extent. The multilayer perceptron artificial neural network (MLPANN) trained by Levenberge–Marquardt algorithm was employed to predict the TCOD, ammonia, phosphate and turbidity of the effluent which resulted in R of 0.8257, 0.6274, 0.7961 and 0.6884, respectively. The optimization of MLPANN by GA performed better than MLPANN with R of 0.8390, 0.7650, 0.8107 and 0.7365 for TCOD, ammonia, phosphate and turbidity, respectively. The B-spline quasi interpolation indicates a more accurate prediction due to its R of 0.9619, 0.8806, 0.8307 and 0.7856 for TCOD, ammonia, phosphate and turbidity, respectively. The B-spline quasi interpolation operation time is notably lower than ANN and ANN-GA. In addition, it has a simple algorithm and is implemented easier than the artificial neural network model.