Comparative study of different wavelet-based neural network models to predict sewage sludge quantity in wastewater treatment plant.

Abstract

In this study, artificial neural networks (ANNs) including feed forward back propagation neural network (FFBP-NN) and the radial basis function neural network (RBF-NN) were applied to predict daily sewage sludge quantity in wastewater treatment plant (WWTP). Daily datasets of sewage sludge have been used to develop the artificial intelligence models. Six mother wavelet (W) functions were employed as a preprocessor in order to increase accuracy level of ANNs. In this way, a 4-day lags were considered as input variables to conduct training and testing stages for the proposed W-ANNs. To compare performance of W-ANNs with traditional ANNs, coefficient of correlation (R), root mean square error (RMSE), mean absolute error (MAE), and Nash-Sutcliffe efficiency coefficient (NSE) were considered. In the case of all wavelet functions, it was found that W-FFBP-NN (R = 0.99 and MAE = 5.78) and W-RBF-NN (R = 0.99 and MAE = 6.69) models had superiority to the FFBP-NN (R = 0.9 and MAE = 21.41) and RBF-NN (R = 0.9 and MAE = 20.1) models. Furthermore, the use of DMeyer function to improve ANNs indicated that W-FFBP-NN (RMSE = 7.76 and NSE = 0.98) and W-RBF-NN (RMSE = 9.35 and NSE = 0.98) approaches stood at the highest level of precision in comparison with other mother wavelet functions used to develop the FFBP-NN and RBF-NN approaches. Overall, this study proved that application of various mother wavelet functions into architecture of ANNs led to increasing accuracy of artificial neural networks for estimation of sewage sludge volume in the WWTP.