Artificial intelligence-based response surface progressive optimality algorithm for operation optimization of multiple hydropower reservoirs

Hydropower reservoir operation is critical to ensuring reliable water and energy supply, supporting sustainable economic and social development. Although the progressive optimality algorithm (POA) is a famous modified dynamic programming technique for resolving multistage decision-making problems, its standard method struggles with poor performance in large-scale multireservoir operation problems due to the dimensionality issue. The computation burden grows exponentially with the increase of state variables, making it challenging to find optimal solutions. To overcome this challenge and improve POA's performance, an effective response surface-based progressive optimality algorithm (RSPOA) is proposed for multireservoir system operation optimization. RSPOA decomposes the original multistage problem into numerous easy-to-solve two-stage subproblems. Additionally, an artificial intelligence-based response surface model is integrated to reduce the huge computation required in determining a modified solution for each subproblem. The simulations show that compared to the standard POA method, RSPOA can make obvious improvements in execution efficiency in various operation scenarios. For instance, in the 4-reservoir system in Wu River with 19 discrete states and dry runoff, RSPOA-LSTM achieves about 79.2 % reductions in the computation time of POA. Thus, RSPOA proves to be an effective tool to solve the complex operation optimization challenges of multireservoir systems.