Abstract: To fully utilize the Pareto frontiers’ information of multi-objective optimization of cascade reservoir, we present a new method for extracting operation rules through optimizing their parameters to approximate successively the operation results toward Pareto frontiers. We develop a multi-objective optimal operation model, solve it using the nondominated sorting genetic algorithm II (NSGA-II), and then obtain the sets of Pareto frontiers for three typical years of wet, normal and low flows. These Pareto frontiers are used as the training samples of a surrogate model to realize a function expression of two-dimensional Pareto frontiers. We construct the basic form of the operation rules using multiple linear configurations, and optimize their parameters used for the typical years through successive approximation to Pareto frontiers. This method avoids destroying the integrity of Pareto frontiers in extraction of multi-objective optimization operation rules due to decision-making, and seek the approximated Pareto solution as the optimized results. It is verified in the case study of the Xiluodu-Xiangjiaba cascade reservoirs. We have extracted successfully the typical years' operation rules for the two goals of power generation and ecology in non-flood season, and verified the rules under similar conditions.

Key words: cascade reservoir operation, multi-objective optimization, operation rules, surrogate model, Pareto frontier