Comparing the performance of the DP, SDP and SSDP models for optimizing the operation of multi-purpose reservoirs (Case study: Zayandeh-Rood dam)

Dynamic Programing models are appropriate tools for deriving optimal operation policies of reservoirs, especially during the water scarcity situation. In this paper, the performances of a variety of these models including deterministic Dynamic Programming (DP), Stochastic Dynamic Programming with inflow classification (SDP_Class), Stochastic Dynamic Programming based on inflow scenarios (SDP_Scenario) and Sampling Stochastic Dynamic Programming (SSDP) were evaluated and compared for Zayandeh-Rood multi-purpose system. For this aim, beside the complete supply of domestic, industrial and environmental demands, optimal operation policies considering constant agricultural demand and then both agricultural and hydropower ones were derived. Finally the performance of optimization models was evaluated and compared based on simulation models, computational load of optimization models as well as some efficiency criterion such as temporal and volumetric reliability, maximum and average shortage. Results showed that stochastic models like SDP and SSDP performed better than DP models, considerably. This superiority was considerable, especially during extreme water shortage conditions. Despite the SSDP preference its computational load as well as its execution time were higher than others. In order to alleviate this problem, pre-allocation and vectorization techniques were applied and they had a significant effect on the run-time reduction of all models, particularly on SSDP.