Quantifying the robustness of optimal reservoir operation for the Xinanjiang-Fuchunjiang Reservoir Cascade

In this research we investigate the robustness of the common implicit stochastic optimization (ISO) method for dam reoperation. As a case study, we focus on the Xinanjiang-Fuchunjiang reservoir cascade in eastern China, for which adapted operating rules were proposed as a means to reduce the impact of climate change and socio-economic developments. The optimizations were based on five different water supply and demand scenarios for the future period from 2011 to 2040. Main uncertainties in the optimization can be traced back to correctness of the assumed supply and demand scenarios and the quality and tuning of the applied optimization algorithm. To investigate the robustness of proposed operation rules, we (1) compare cross-scenario performance of all obtained Pareto-optimal rulesets and (2) investigate whether different metaheuristic optimization algorithms lead to the same results. For the latter we compare the originally used genetic algorithm (Nondominated Sorting Genetic Algorithm II, NSGA-II) with a particle swarm optimization algorithm (MOPSO). Reservoir performance was measured using the shortage index (SI) and mean annual energy production (MAEP) as main indicators. It is found that optimal operating rules, tailored to a specific scenario, deliver at most 2.4% less hydropower when applied to a different scenario, while the SI increases at most with 0.28. NSGA-II and MOPSO are shown to yield approximately the same Pareto-front for all scenarios, even though small differences can be observed