Journal of Engineering, Project, and Production Management, 2026, 16(7), 2026-0052
Graph Neural Network Collaborative Optimization of Multi-Agent Cascade Hydropower Unit Combination under Spot Market Environment
1 Senior
Engineer, State Grid Sichuan Electric Power Company, China, E-mail:
xiaoc0013@sc.sgcc.com.cn
(corresponding author).
2 Senior Engineer, State Grid
Sichuan Electric Power Company, China
3 Senior Engineer, State Grid
Sichuan Electric Power Company, China
4 Senior Engineer, State Grid
Sichuan Electric Power Company, China
5 Senior Engineer, State Grid
Sichuan Electric Power Company, China
6 Senior Engineer, Sichuan Power
Exchange Center Co., Ltd., China
7 Senior Engineer, Sichuan Power
Exchange Center Co., Ltd., China
8 Senior Engineer, State Grid
Sichuan Electric Power Company Economic Research Institute, Chengdu
610041, China
Project Management
Received May 8, 2026; revised June 18, 2026; accepted July 29, 2026
Available online August 12, 2026
Abstract: With the growing frequency of spot market transactions, cascade hydropower units are required to make rapid decisions in complex hydrological conditions, uncertain loads, and multi-agent strategic games. Traditional scheduling methods fall short in capturing the spatiotemporal coupling of cascade reservoirs and the high-dimensional structural information arising from the dynamic strategic feedback among multiple agents. To address this, this paper applies a graph neural network collaborative optimization framework with an enhanced multi-agent structure. First, a cascade hydropower unit combination relationship graph is constructed, and topological encoding is applied to features such as reservoir capacity, output, bidding price, and bidding characteristics for wind, solar, hydro, thermal, and storage. An interactive graph attention mechanism is designed to strengthen the influence of upstream and downstream and agent strategies. Subsequently, a time-series graph is used to embed a unified representation of the dynamic coupling features among price, hydrological conditions, and strategies, capturing the impact of wind and solar fluctuations, energy storage charging and discharging behavior, and changes in the marginal cost of thermal power on the clearing results. Furthermore, an end-to-end differentiable Graph Neural Network (GNN) fusion module for scheduling optimization is constructed, directly embedding the graph model output into cascade constraint optimization to achieve a joint solution of bidding clearing and cascade scheduling. Ultimately, a multi-objective collaborative learning strategy is employed to improve hydropower utilization, ensure compliance with safety constraints, and enhance multi-agent benefits, enabling the model to achieve higher collaborative decision-making capabilities and generalization performance in the spot market. Experiments show that the satisfaction rate of key constraints under typical operating conditions is above 0.8, and the hydropower bidding collaboration reaches 0.88 with a price fluctuation of only 2.3%, significantly better than wind power's 12.5%; the hydropower utilization efficiency during the normal water season reaches 92.8%, with a water abandonment rate as low as 1.2%; the safety margin in normal scenarios exceeds 15%, with a maximum safety score of 0.88, verifying the model's comprehensive advantages in collaborative decision-making, hydropower optimization, and safe operation.
Keywords: Spot market; cascaded hydropower; unit commitment optimization; graph neural network; collaborative optimization framework.
Copyright © Journal of Engineering, Project, and Production Management (EPPM-Journal).
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Requests for reprints and permissions at eppm.journal@gmail.com.
Citation: Xiao, C., Miao, S., Zhang, C., Li, M., Tang, L., Wang, J., Peng, L., and Hou, Y. (2026). Graph Neural Network Collaborative Optimization of Multi-Agent Cascade Hydropower Unit Combination under Spot Market Environment. Journal of Engineering, Project, and Production Management, 16(7), 2026-0052.
