Estimating effluent turbidity in the drinking water flocculation process with an improved random forest model

Abstract

Abstract During drinking water treatment, the uncertain changes of raw water quality bring great difficulties to the control of flocculant dosage, especially because the feedback information based on the effluent turbidimeter of the sedimentation tank can only be obtained after a long time when the influent water quality changes due to the large lag characteristics of the flocculation process. Prediction of effluent turbidity of the sedimentation tank can effectively solve the aforementioned problem. Given that it is difficult for the ordinary random forest (RF) model to accurately predict the effluent turbidity of a sedimentation tank for complicated changes of raw water quality, an improved random forest (IRF) model composed of long-term and short-term parts is proposed, which can capture the periodicity and time-varying characteristics of influent water quality data. The experimental results show that the root mean square error and mean absolute percentage error of IRF model in Baiyangwan waterworks are improved 67.52% and 67.91% respectively, compared with those of the ordinary RF model. The proposed effluent turbidity predictions are also successfully developed in Xujiang waterworks and Xiangcheng waterworks of Suzhou, China. This research provides an effective method for real-time prediction of the effluent turbidity of sedimentation tank according to the influent water quality data.