New Constraint-Handling Technique for Evolutionary Optimization of Reservoir Operation

Abstract

Evolutionary optimization of reservoir operation is subject to complex physical and operational constraints. Constraint-handling techniques (CHTs) in this field are predominantly problem-specific or based on certain evolutionary algorithms; generally applicable CHTs are seldom tested against reservoir scheduling problems. This study proposes an independent CHT to accommodate the reservoir operation constraints, called the nondomination rank-based adaptive method (NRAM). The NRAM is straightforward to use and free of parameter tuning. The process emphasizes exploiting information from infeasible individuals and preserving them to promote convergence to global optima on a feasible space boundary. Moreover, the method adjusts the population composition dynamically to facilitate exploration or local search. The NRAM was applied to the hydropower scheduling of the Three Gorges Reservoir and Gezhouba Reservoir in China. Results show that the NRAM performs slightly better than three other well-regarded CHTs but requires mildly longer computational time. In addition, the genetic algorithm with the NRAM outperforms dynamic programming that is commonly used for hydropower scheduling. The operation schedules the NRAM provides are well suited for maximizing hydropower generation with all constraints satisfied.