Fig. 5: Comprehensive evaluation of ML-driven dosing programs.

a Real incoming and outgoing water and stage turbidity of the water plant during the experimental validation.The moments corresponding to the dark gray background do not have laboratory validation, and the moments corresponding to the light gray background have laboratory validation.b Schematic diagram of a scaled-down flocculation tank.c Changes in turbidity under the lab using the original method of dosing at the water plant, and dosing using the ML method.d Turbidity during the actual operation of the water plant, and simulation of the turbidity in the application of ML.e Change in total daily PAC dosing. f Change in total treatment cost per unit volume of wastewater. g Change in percentage of total cost for multiple chemicals. h For each feature, 10% of the values are randomly selected and set to 0, and 100 Monte Carlo simulations are performed. i For key features, randomly select 10%, 20%, 30%, 40%, and 50% of the values and set them to 0, then perform 100 Monte Carlo simulations. TUR: turbidity. PAC: polybasic aluminum chloride.