WQI-Based Algorithmic Decision-Making for Water Withdrawal Depth in Drinking Water Reservoirs

Abstract

Determining optimum withdrawal depth for selective water withdrawal is crucial for managing the water quality of withdrawn water or reservoir. In this paper, we consider non-stratified periods, relatively less studied, and develop an algorithmic decision-making system based on water quality index. Our approach takes three inputs (depth profile data, current withdrawal depth, and weights in water quality index) and determines the water withdrawal depth. Finding the weights associated with these parameters in the index is formulated as an optimization problem in which the matching between our recommended depth and operator's decision is maximized. Historical operator decisions over a 14-year period from a real water treatment plant were used to optimize these weights. We utilized surrogate optimization, genetic algorithm, pattern search, and brute-force solvers for this optimization problem. Brute-force solver achieved a 93.8% matching accuracy in predicting the water withdrawal depth. As the other three solvers require an initial solution, which can influence the quality of their results, the brute-force solver was selected as the most reliable approach. Our approach accurately predicts the decisions made by water treatment plant operators regarding water withdrawal from a reservoir with a multi-level intake structure during periods without thermal stratification.