A novel framework of deriving joint impoundment rules for large-scale reservoir system based on a classification-aggregation-decomposition approach

Abstract. Joint and optimal impoundment operation of the large-scale reservoir system has become more crucial for modern water management. Since the existing techniques fail to optimize the large-scale multi-objective impoundment operation due to the complex inflow stochasticity and high dimensionality, we develop a novel combination of parameter simulation optimization and classification-aggregation-decomposition approach here to overcome these obstacles. There are four main steps involved in our proposed framework: (1) reservoirs classification based on geographical location and flood prevention targets; (2) assumption of a hypothetical single reservoir in the same pool; (3) the derivation of the initial impoundment policies by the non-dominated sorting genetic algorithm-II (NSGA-II); (4) further improvement of the impoundment policies via Parallel Progressive Optimization Algorithm (PPOA). The framework potential is performed on China's mixed 30-reservoir system in the upper Yangtze River. Results indicate that our method can provide a series of schemes to refer to different flood event scenarios. The best scheme outperforms the conventional operating rule, as it increases impoundment efficiency from 89.50 % to 94.16 % and hydropower generation by 7.70 billion kWh (or increase 3.79 %) while flood control risk is less than 0.06.