Many-Objective Operation of Selective Withdrawal Reservoirs Including Water Quality Targets

Current water management practices are mostly targeted toward quantitative uses, while water quality is usually addressed separately as an independent problem. In this work, a novel reinforcement learning approach is proposed for designing Pareto-optimal operating policies for reservoirs equipped with selective withdrawal systems (SWSs). This approach exploits the feedback between the selection of outlet locations for water quantity demands and the water quality patterns within the reservoir, making it possible to satisfy downstream objectives and, simultaneously, affect in-reservoir water quality. Dealing with many water interests, a novel method is presented for reducing the number of objectives via Non-negative Sparse Principal Component Analysis (NSPCA). The approach is demonstrated on the real-world case study of Tono Dam (Japan). Results indicate that potential control over water quantity and quality can be gained by effectively exploiting SWS's flexibility, taking advantage of the NSPCA-based reduction of the operating objectives.