Short-term water quality prediction of reclaimed water plant effluent and key measurement sections based on a surrogate prediction model.

Abstract

Reclaimed water serves as the secondary water source for cities. To ensure the sustainable utilization of reclaimed water, it is necessary to establish reliable short-term water quality prediction for pollutant discharge from reclaimed water plants and at key sections along rivers. The MIKE 11 model was constructed to assess the influence of pollutant discharge from each reclaimed water plant on the water quality of the key section along the river, which reveals that the discharge from the XZ I reclaimed water plant had the most significant negative impact on the river water quality of the key section. Based on the simulation data from the MIKE 11 model, an XGBoost surrogate model was developed to map the daily pollutant discharge data from the reclaimed water plants to the concentrations of three pollutants (COD, NH3N, and TP) at the key section. The RMSE values of the predicted pollutant concentrations of each reclaimed water plant were all lower. Given the substantial impact of the XZ I plant's discharges on the key section's water quality, a VMD-CPO-LSTM prediction model was developed using the plant's daily pollutant discharge data to achieve short-term water quality forecasts. This model was then transferred and applied to other reclaimed water plants to validate the applicability of this model in the Nanming River Basin. The RMSE values were less than 1 and the R² values were all higher than 0.8 compared to the measured values. Finally, short-term water quality prediction at key sections downstream was realized by coupling the surrogate model and the VMD-CPO-LSTM prediction model. According to the results of the appraisal, the prediction error of the surrogate model for three pollutants at key sections was lower than 12 % compared to the measured values. In addition, results show that the calculation time of the surrogate prediction model was 2.92 % of that of the MIKE 11 model, and the calculation errors between the MIKE 11 model and the surrogate prediction model differed by less than 3 %. The research results can guarantee the efficient utilization of reclaimed water resources and promote the green development of cities.