Multi-Scale Fuzzy Inference System for Influent Characteristic Prediction of Wastewater Treatment

Accurate influent characteristic prediction is vital to maintain the stable performance of wastewater treatment processes. In this work, an associated approach based on the wavelet packet decomposition (WPD) and adaptive network-based fuzzy inference system (ANFIS) is proposed to address this issue. In this method, the WPD is first adopted to decompose the historical data of the influent characteristic into wavelet coefficients in different scales. The time sub-series, which are obtained with a single branch reconstruction of the wavelet coefficients in each scale, are then utilized to build the ANFIS regression model. The predicted sub-results in each scale are finally summarized into an eventual predicted result. Moreover, a particle swarm optimization (PSO) algorithm is employed to acquire the optimal parameters of the multi-scale ANFIS, and chaos theory is utilized to determine the input variables of the multi-scale ANFIS. The reported approach is investigated by the influent characteristic data, including the chemical oxygen and biochemical oxygen demands from a wastewater treatment plant (WTP) in southwest of China. Two peer models are introduced for a comparison study. The results show that the developed approach has superior performance in terms of the mean absolute error (3.346 and 1.384), mean absolute percentage error (1.804% and 1.800%), root mean square error (3.988 and 1.788), and correlation coefficient (0.960 and 0.964), and can accurately predict the influent characteristic of the WTP.