Hybrid data and knowledge driven approach for determining coagulant dosing in drinking water treatment plants

Abstract

The large time-delay in the coagulation process at drinking water treatment plants complicates accurate coagulant dosage determination. In this study, we proposed a Gated Recurrent Unit model enhanced with a local attention mechanism (GRU_LA) to precisely predict the required coagulant dosage and effluent turbidity. These models were integrated into a feed-forward-feedback composite control strategy, forming a data-driven control for coagulant dosing in drinking water treatment plants. Additionally, a hybrid rule-based expert system was also proposed as a knowledge-driven control component and combined with data-driven control to achieve a coagulant dosing system. Experimental results demonstrated that GRU_LA more effectively predicted the turbidity of effluent from the coagulant dosage, achieving a Mean Absolute Percentage Error (MAPE) of 1.61% for coagulant dosage and 0.86% for effluent turbidity, with a coefficient of determination (R²) of 0.90 and 0.94, respectively. After implementing the coagulant dosing control system in a drinking water treatment plant, the coefficient of variation of effluent turbidity throughout 2023 decreased by 5.58% compared to that of the monthly average in 2021, and the average annual coagulant usage was reduced by 7.83 mg L⁻¹, marking a 27.96% decrease and significantly lowering the cost of coagulants.