关键词: 短期优化, 多目标调度, 数据挖掘, 典型出力提取, 计划执行完成率

Abstract:

[Objective] Short-term optimal scheduling of cascade reservoirs is complicated by hydraulic connections among power stations and environmental factors, and most studies neglect the risk of deviating from normal operations. Therefore, this paper proposes a plan completion rate indicator that comprehensively measures deviation and scheduling target satisfaction using the Feature Similarity Index (FSI), and constructs a short-term multi-objective optimization model aiming to maximize cascade power generation and plan completion rate. [Methods] For the short-term multi-objective joint optimization scheduling of the lower-Jinsha River cascade, K-means clustering of historical plant data was first conducted. Silhouette coefficient (SC) and Davies-Bouldin index (DBI) were used to select the optimal number of clusters for each station on a monthly basis. Typical generation patterns were then extracted based on these optimal clusters to provide a data foundation for the subsequent scheduling model. Further, a multi-objective model maximising power generation and plan-completion rate, using the Feature Similarity Index (FSI) to comprehensively evaluate deviation and the degree of scheduling target fulfillment, subject to water balance, water level, discharge and output constraints, was built and solved by NSGA-II (using a water-level corridor to handle hard constraints). The final scheme was selected using the entropy weight method and the VIKOR multi-criteria compromise ranking method. [Results] In the case study, Pareto solutions showed a clear trade-off between power generation (921.52-922.13 million kW·h, range 0.61 million kW·h) and FSI (0.831 6-0.872 6, range 0.041) with evenly distributed results. Entropy weighting assigned weights of 0.513 2 to generation and 0.486 8 to FSI. The VIKOR-selected compromise scheme (closeness coefficient 0.002 2) yielded 921.98 million kW·h (7.4% above normal 857.2 million kW·h) and an FSI of 0.861 2, occupying 74.63% and 72.03% of their respective ranges. Except for Xiluodu, the outputs variation trend of Xiangjiaba, Three Gorges, and Gezhouba plants generally aligned with the typical generation patterns. [Conclusion] The results show that the model effectively balances power benefits with output process deviation and quantifies generation improvements under varying plan completion rates. By scientifically selecting optimal cascade scheduling schemes, it offers reliable references for operators of the cascade reservoirs in the lower-Jinsha River-Three Gorges region and supporting medium-/long-term schedule adjustments as well as the formulation of reservation-period plans.

Key words: short-term optimization, multi-objective scheduling, data mining, typical output extraction, plan completion rate