Effective reservoir operation in semi-arid regions is challenging due to erratic rainfall, variable inflows, high evaporation losses, and competing demands for limited water resources, particularly irrigation and domestic use. To address these issues, this study evaluates three advanced metaheuristic optimization algorithms, Rao-4, Jaya, and TLBO (Teaching - Learning Based Optimization), for deriving optimal operational policies for the Dharoi Reservoir, India. These algorithms were selected for their parameter-free nature, which simplifies implementation and avoids the sensitivity to tuning required by traditional methods such as Genetic Algorithms (GA). The newly introduced Rao-4 algorithm was first validated using the continuous type four reservoir (CTFR) benchmark problem and compared with TLBO and Jaya through the TOPSIS approach. Results showed TLBO ranked highest, followed by Rao-4, which achieved competitive performance with the lowest computational time. The algorithms were then applied to Dharoi Reservoir under five dependable inflow conditions (50, 55, 60, 70, and 75%). TOPSIS evaluation indicated that TLBO achieved the best performance across most inflow conditions, while Rao-4 ranked first at 50% dependability. Optimal-release policies for 50% and 55% inflows were lower than actual releases, suggesting opportunities to expand irrigation or divert water to deficit regions, though with reduced reliability.