Modelling coagulant dosage in drinking water treatment plant using hybrid machine learning based on empirical wavelet transform

The coagulation process is one of the most important steps in drinking water treatment plant and its accurate determination plays a crucial role. In this study, simple and hybrid artificial intelligence models were applied for better prediction of coagulant dosage rate in Algeria drinking water treatment plant. First, the multilayer perceptron neural network (MLPNN), and random forest regression (RFR) were applied for predicting the coagulant dosage rate, using six raw water quality variables, these include dissolved oxygen (DO), turbidity (TU), temperature (TE), conductivity (SC), the pH of water, and ultraviolet absorbance (UV). From the obtained results, it was found that coagulant dosage rate is highly difficult to estimate by single models and the research should be oriented toward the development of a new modelling strategy. Second, the six input variables were further decomposed using the empirical wavelet transform (EWT) algorithm, leading to the formation of an ensemble of new variables called multiresolution analysis (MRA) which were combined and used as new input variables. Our hybrid models based on EWT guaranteed significant improvement compared to the single models. The results showed that the MLPNN-EWT model reduced the errors very much, and they greatly enhanced the fitting capability when compared to the single models, exhibiting a Pearson correlation coefficient (R), Nash-Sutcliffe efficiency (NSE), root-mean-square error (RMSE), and mean absolute error (MAE) of approximately ≈0.935, ≈0.901, ≈2.812, and ≈1.923. These improvements in coagulant dosage prediction are consistent and robust.