Improving Flood Control Optimization Scheduling for Cascade Reservoirs Considering Various Runoff Scenarios

Abstract

Cascade reservoirs develop flood control dispatch plans to address potential flood disasters. However, runoff is an uncertain natural process that can seriously affect the actual effectiveness of dispatch plans. Therefore, it is essential to consider various possible runoff scenarios when formulating flood control dispatch plans. This paper uses a time-series generative adversarial network (timeGAN) to generate a large number of possible runoff samples and construct a comprehensive flood control optimization scheduling model for cascade reservoirs. This model is then solved using the distributionally robust optimization algorithm to produce the flood control scheduling plan under the worst possible runoff scenario. First, the runoff generation model based on timeGAN was constructed, generating a large number of possible runoff sequence samples that could occur in reality. Second, consider the three objectives of flood control dispatch comprehensively: flood risk; water storage; and water resource utilization. Then, combine these objectives with flood control dispatch constraints to construct the flood control optimization scheduling model for cascade reservoirs. Third, a data-driven Wasserstein runoff fuzzy set is constructed using a large number of generated runoff sequence samples. The distributionally robust optimization method is then used to solve the constructed cascade reservoir flood control optimization scheduling model, thereby obtaining the flood control scheduling plan under the worst runoff sequence distribution within the runoff fuzzy set. Finally, the cascade reservoirs in the lower reaches of the Jinsha River in China were selected as the research object. The results of the calculation example demonstrate that: 1) The maximum incoming water in the generated Wudongde inflow sample increased by 20.3% compared to the historical sample, while the minimum value decreased by 36.9%, greatly enriching the runoff sample; 2) simulation scheduling results show that the flood control dispatch plan effectively reduced the peak outflow of Xiangjiaba by 15% and completed the water storage task; and 3) compared with stochastic programming, the proposed method increases the constraint satisfaction probability by 4.3% at the cost of a 0.55% increase in the maximum outflow. Compared with robust optimization, it reduces the maximum outflow by 4.0% while incurring only a 4.9% decrease in the constraint satisfaction rate. Overall, the proposed method effectively balances the risk and benefit in cascade reservoir flood control under runoff uncertainty.