In this paper, the economic dispatch system based on neural network optimization algorithm is designed for the cascade hydropower regional centralized power generation scenario. The system integrates the functions of hydrologic monitoring, runoff prediction, medium and long-term optimization scheduling, annual monthly weekly planning, working condition recognition, early warning analysis and feedback writeback, and jointly codes the water level, flow, storage capacity, output, maintenance plan and power grid command of three power stations of Jiangping River, Bashou River and Shuojinshan. And the rolling solution is completed under the constraints of water balance, minimum lower discharge, water level boundary and unit operating conditions. The results show that the average flow prediction error of ten days is 6.8%, the monthly planning deviation is 3.5%, the hydropower utilization rate is 91.8%, and the average scheduling response delay is 1.7 s, which can support plan generation, online correction and closed-loop scheduling execution. At the same time, the system identifies the water state of abundance and dryness according to the frequency threshold of 30% and 70%, and the overall operation is stable, and the plan connection is smooth.
