Load Optimization Distribution Among Cascade Hydropower Stations Based on The A* Algorithm

Abstract

The growing integration of renewable energy has intensified peak-shaving and frequency-regulation demands on hydropower systems, causing frequent load fluctuations and imposing stringent timeliness requirements on load distribution among cascade hydropower plants (LDCHP). Conventional methods such as dynamic programming (DP) face prohibitive computational costs, while intelligent optimization algorithms may suffer from instability and parameter sensitivity. To address this challenge, this study formulates LDCHP as a finite multi-stage state-space search problem, with station operating states represented as search nodes. A tailored A* algorithm is developed with three key components: an actual cost function based on energy-storage consumption induced by generation decisions along the searched path, a heuristic function based on an aggregated hydropower plant model to efficiently estimate remaining cost, and an adjacent-node expansion procedure enforcing operational constraints. The method is evaluated on a 96-period (15-min intervals) intra-day LDCHP problem using three cascade stations in the middle reaches of the Dadu River Basin, China, as a representative case study. Under a 50 m3/s discretization step, the A* algorithm achieves an average speedup of approximately 5,347 times compared with DP, while the cascade energy-storage consumption obtained by the A* algorithm is only 1.65% higher than that obtained by the DP algorithm under the same total generation requirement. These results suggest that the proposed method provides an efficient decision-support tool for short-term load distribution in cascade hydropower systems with similar dispatching characteristics.