Many-Objective Direct Policy Search in the Dez and Karoun Multireservoir System, Iran

Abstract

In this study we propose a novel approach to design Pareto-optimal operating policies via many-objective direct policy search in the Dez and Karoun multireservoir system (Iran), where the dimension of the system, the dry climate, and the presence of competing demands pose a number of challenges to water planners and managers. The three power plants connected to the main reservoirs in the modeled system (i.e., Dez, Karoun, and Masjed Soleyman) account for 20% of the national hydropower generation capacity. Irrigation and domestic supply, especially to the city of Ahwaz, are also strategic objectives, along with flood protection downstream of the dams. Given the complexity of the Dez and Karoun system and the multiple interests involved, the design of Pareto-optimal operating policies via many-objective direct policy search, combined with their a posteriori evaluation, represents an effective tool to support sustainable water reservoirs management in Iran. Our preliminary results show that the proposed decision analytic framework allowed the identification of the set of Pareto-optimal policies and discovered key tradeoffs to eventually aid the selection of few candidate compromise solutions that balance the competing objectives.

Multiobjective Direct Policy Search Using Physically Based Operating Rules in Multireservoir Systems

“…Controlling a network of five storages using such a low number of decision variables significantly reduces computational complexity compared to prior DPS approaches. For instance, Biglarbeigi et al (2014) used 117 parameters to optimize the release policy of a network of three reservoirs. Giuliani et al (2018) used 176, and 134 parameters for a system consisting of four reservoirs.…”

“…The direct policy search expresses a release policy using a given family of parameterized functions (Deisenroth et al 2013). It was also extended to multiobjective problems (Biglarbeigi et al 2014) and MOEAs were used to solve it. For instance, Giuliani et al (2016) found radial basis functions (RBF) to be particularly useful for this task.…”

Partitioning the Impacts of Streamflow and Evaporation Uncertainty on the Operations of Multipurpose Reservoirs in Arid Regions

“…Each optimization was run for 100,000 function evaluations. The analysis of runtime progress of the optimization(Biglarbeigi 2014) ensures the attainment of high fidelity approximations of the Pareto approximate set with this number of function evaluations. To improve solution diversity and avoid dependence on randomness, we run 20 independent optimization trials.…”