This study aims compares how different formulations of a reservoir operation problem with conflicting objectives affect the quality of the generated solution set. Six models were developed for comparative analysis: three using dynamic programming (DP) and three using the evolutionary multi-objective direct policy search (EMODPS) algorithm. Afterward, to improve the quality of the generated solution set, an EMODPS model was selected and coupled with zone-based hedging policy that is currently being applied in real-world reservoir operations. The solutions generated by each model were then evaluated regarding proximity to the ideal and three eminent performance indices (risk, resiliency, and vulnerability). The proposed methodology was applied to a multi-purpose reservoir located in South Korea, Boryeong Dam, which had suffered a multi-year drought recently. Consequently, the solution sets from the EMODPS model yielded closer results than those of the stochastic DP model for optimality and diversity. Although the solutions from the algorithm performed better than actual operation results under normal conditions, the actual operations executed based on the zone-based hedging rule outperformed the other two in case of droughts. Among the EMODPS models, one with the fewest parameters, the EMODPS-Gaussian model, resulted in better solutions for all cases. Finally, coupling the real-world policy with the optimally derived solutions in the case of droughts improved the frequency, duration, and magnitude of the water supplies whereas the water users experienced an improvement in scale at the expense of more recurrent failures.