Successful prediction for coagulant dosage and effluent turbidity of a coagulation process in a drinking water treatment plant based on the Elman neural network and random forest models

Abstract

The uncertainty of the changes in the quality of raw water, and the long lag in the process of coagulation introduce significant difficulties in eliminating turbidity during the treatment of drinking water. In this study, the Elman neural network (ENN) and the random forest (RF) models are proposed to accurately predict the required coagulant dosage and turbidity of the effluent. The experimental results showed that the ENN better predicted the coagulant dosage, with values of the mean absolute percentage error (MAPE) in the range of 2.38–5.73%, whereas the RF model was preferable for predicting the turbidity of the effluent, with values of the MAPE in the range of 6.45–12.52%. The two best-performing models for these two tasks recorded MAPEs of 2.38% and 6.45%. The prediction residuals for the best performance of the ENN model were in the range of −1.0 to 1.0 mg L⁻¹, and the residual of predictions of the best RF model for effluent turbidity was in the range of −0.2 to 0.1 NTU. Thus, these models stabilized the residuals of the turbidities of the effluent and the target within 0.2 NTU, and reduced the cost of coagulant used in the water plant by 10%.