Utilization of WGEP and WDT Models by Wavelet Denoising to Predict Water Quality Parameters in Rivers

Abstract

In this study, new methods based on integrating discrete wavelet transforms (DWT) into artificial neural network (ANN), gene expression programming (GEP), and decision tree (DP) approaches for several applications of water quality index estimation are proposed. The 3-year daily data used in this study, including turbidity (Tur), pH, dissolved oxygen (DO), discharge, and temperature, were measured from the Blue River at Kenneth Road, Overland Park, Kansas, in Johnson County. In addition to the time delays concerning each of the parameters DO, Tur, and pH, the temperature and discharge were considered effective in the climate of the region. The results showed that using wavelets significantly improved the performance of the ANN, DT, and GEP models, particularly in the case of extreme values. The results are comparable and suggest that wavelet-AI conjunction models could be explored as an alternative tool for water quality prediction. The performance of wavelet-gene expression programming (WGEP), which was moderately better than wavelet-artificial neural network (WANN) and wavelet-decision tree (WDT), is very promising and hence supports the use of WGEP in predicting river quality data. The results showed that the WGEP model decreased the mean absolute percentage error for the WDT, WANN, GEP, ANN, and DT models from .18, .33, .41, .47, and .35 mg/L, respectively, to .17 mg/L for the DO index, and from .26, .18, .36, .8, and .51 mg/L to .4 mg/L for the pH index, respectively. The WANN model also dropped the mean absolute percentage error for the WDT, WGEP, GEP, ANN, and DT models from 9.72, 6.68, 13.98, 8.88, and 14.81 FNU to 5.6 FNU for the Tur index. In this study, hybrid models provided more precise predictions for extremely high values.