A Genetic-Algorithm-Based Multi-Objective Decision Support System for Cascade Reservoir Water Resource Management

Abstract

Growing volumes of publicly available hydrological, meteorological, and reservoir-operation data create new opportunities for evidence-based water resource management. However, converting these data into actionable operating strategies requires optimisation frameworks capable of addressing the non-linear, non-convex, and conflicting objectives of reservoir operation. This study develops and evaluates a genetic-algorithm-based decision support system (DSS) for a three-reservoir cascade serving irrigation, municipal water supply, hydropower generation, and flood control. A real-coded, constraint-handling Non-dominated Sorting Genetic Algorithm II (NSGA-II) is implemented in Python and coupled with a mass-balance simulation model driven by a ten-year stochastically generated monthly inflow record calibrated to a monsoon-influenced river basin. The optimisation considers a 45-parameter cyclic hedging-rule policy, producing implementable and repeatable management strategies rather than fixed release schedules. Eight independent runs (150 individuals, 300 generations) produce a consolidated Pareto front of 417 non-dominated policies balancing water-supply deficit, hydropower generation, and flood-control performance. A TOPSIS-based compromise-selection layer converts these solutions into operating recommendations under three managerial preference profiles. Compared with a conventional Standard Operating Policy baseline, the balanced GA-optimised policy increases hydropower generation from 6,453.9 to 6,564.2 GWh over ten years and reduces the flood-control exceedance index by 90%, while the water-supply-priority policy increases generation to 6,789.3 GWh. The results demonstrate a deliberate reliability-resilience-vulnerability trade-off, with GA policies rationing water more frequently but moderately, thereby redistributing risk across the cascade. Sensitivity and convergence analyses further confirm the robustness of the optimisation framework.