This study presents an integrated framework for the simulation and optimization of water quality in the Ekbatan Reservoir, aimed at minimizing total dissolved solids (TDS) concentration. The proposed framework combines the CE-QUAL-W2 model, an XGBoost-based surrogate model, and two metaheuristic optimization algorithms, namely the Enhanced Vibrating Particle System (EVPS) and Particle Swarm Optimization (PSO). First, the CE-QUAL-W2 was calibrated and validated using observed field data. The model successfully reproduced the hydrodynamic and water quality dynamics of the reservoir, achieving Nash-Sutcliffe Efficiency (NSE) values of 0.97 and 0.98 for reservoir storage and water surface elevation, respectively. The simulation outputs were subsequently used to train the XGBoost surrogate model. The surrogate model demonstrated satisfactory predictive capability, with R² = 0.9176, MAE = 1.7012, and RMSE = 2.5133, confirming its ability to accurately estimate water quality variables. The trained model was then embedded within the optimization framework to evaluate reservoir operating policies efficiently. Comparative results showed that EVPS significantly outperformed PSO, achieving an objective function value of 0.2340 compared with 88.3333, although it required slightly longer computational time (332.21 s versus 238.79 s). Furthermore, the operating policies obtained by EVPS produced lower TDS concentrations and more stable released-water quality than those generated by PSO. Overall, the proposed CE-QUAL-W2-XGBoost-EVPS framework effectively reduces computational cost while maintaining high predictive accuracy and improving optimization performance, demonstrating its potential as an efficient decision-support tool for intelligent and sustainable reservoir water quality management.