Water reservoirs quality management using meta-heuristic Algorithms: Analysis and optimization of water quality considering uncertainties

Abstract

Managing reservoir water quality under uncertainty remains a critical challenge in contemporary water resource management. This study introduces a robust simulation–optimization meta-model framework to enhance reservoir outflow quality, focusing on minimizing Total Dissolved Solids (TDS) concentrations. To circumvent the computational limitations of high-fidelity simulators, a Supervised Learning (SL) surrogate model was developed as a substitute for the CE-QUAL-W2 simulator. Achieving a prediction accuracy of 85 %, the SL model effectively captures complex, nonlinear interactions within water quality dynamics. Two hybrid metaheuristic frameworks—Particle Swarm Optimization integrated with SL (PSO-SL) and Enhanced Vibrating Particle System integrated with SL (EVPS-SL)—were implemented to optimize reservoir outflows under uncertainty. Both approaches successfully balanced the competing objectives of meeting downstream water demand and minimizing TDS concentrations, while significantly reducing computational costs and improving convergence behavior. The rigorously calibrated CE-QUAL-W2 model demonstrated high validation scores (NSE = 0.99 for storage volume and 1.00 for water level; PBIAS = −0.05 % and −0.0004 %, respectively), confirming its reliability for surrogate training. Additionally, the study examined uncertainty propagation using two widely adopted sampling techniques: Monte Carlo Simulation and Latin Hypercube Sampling (LHS). Optimization outcomes were assessed using performance metrics—reliability, vulnerability, and resilience. The PSO-SL model, coupled with Monte Carlo sampling, exhibited the most balanced performance, achieving 41 % reliability and 26 % vulnerability. In contrast, EVPS-SL with LHS demonstrated faster convergence (30 % reduction in computational time) but yielded lower reliability (16 %) and higher vulnerability (87 %). This research not only advances reservoir water quality management under uncertainty but also contributes methodologically to the integration of data-driven surrogates and optimization within environmental systems modeling.