A parallel approximate evaluation-based model for multi-objective operation optimization of reservoir group

Reservoir operation optimization can boost the efficiency of water resources utilization, but sometimes has huge search space and time-consuming calculation. Approximate evaluation is one of the mainstream methods to assist evolutionary algorithms to efficiently solve such problems. However, most approximation techniques have to constantly correct accuracy during optimization because of the inability to precisely control approximation errors, resulting in a decrease in computational efficiency. Therefore, by fully mining operating information and deeply integrating function evaluation with mutation operator, this study proposes a novel parallel approximate evaluation-based model (PAEM) to enhance search ability and shorten calculation time as well as realizing accurate control of approximation errors, and establishes a multi-objective operation model PAEM-LSTM by combining PAEM and long short-term memory neural network (LSTM) for the fast formulation of operating rule. The results indicate that: (1) under the same parallelization, compared with three multi-objective evolutionary algorithms and two surrogate-based multi-objective algorithms, PAEM provides significantly better Pareto-optimal solutions at a faster speed (e.g. 32 times faster than NSGA-II) while maintaining extremely low approximation errors; (2) small population size and large mutation size are recommended in PAEM, and moreover, the larger the scale of reservoir group, the higher the computational efficiency of PAEM; and (3) compared with conventional operating rule, the operating rule of NSGAII-LSTM increases hydropower generation by 3.45% and reduces ecological water shortage by 29.74%, while the rule of PAEM-LSTM increases hydropower generation by 3.63% and reduces ecological water shortage by 36.74%. This study sheds a new idea for multi-objective operation optimization.