引用本文: | 王 新,张 良,任晓龙,等.融合图神经网络模型与强化学习的综合能源系统优化调度[J].电力系统保护与控制,2023,51(24):102-110.[点击复制] |
WANG Xin,ZHANG Liang,REN Xiaolong,et al.Optimal scheduling of integrated energy systems by fusing a graph neural network model and reinforcement learning[J].Power System Protection and Control,2023,51(24):102-110[点击复制] |
|
摘要: |
随着人工智能技术特别是强化学习在能源优化调度领域的深入研究,将系统状态表示为向量用于学习的模式,其训练效率与信息利用率较低。针对这一问题,提出了一种融合图神经网络模型与强化学习的综合能源系统优化调度方法。首先,将电-热-气综合能源系统建模为图结构数据,充分利用系统的拓扑信息。其次,提出了基于图神经网络架构的强化学习模型,使其可以充分利用图结构信息实现更快的训练速度,获得更大的探索空间。最后,将表示系统状态的图结构信息送入该模型进行训练,算例仿真验证了该方法的训练效率与探索能力。 |
关键词: 电-热-气综合能源系统 优化调度 深度强化学习 图神经网络模型 |
DOI:10.19783/j.cnki.pspc.230381 |
投稿时间:2023-04-08修订日期:2023-06-13 |
基金项目:国家自然科学基金项目资助(62176227,U2066213);中央高校基本科研业务费项目资助(20720210047);国家电网陕西省电力公司科技项目资助(SGSNXT00GCJS2200106) |
|
Optimal scheduling of integrated energy systems by fusing a graph neural network model and reinforcement learning |
WANG Xin1,ZHANG Liang2,REN Xiaolong1,ZENG Yizhou2,SI Hengbin1,CHEN Xi1,YANG Le1,ZHANG Zhihong2 |
(1. State Grid Shaanxi Information and Telecommunication Company, Xi’an 710065, China;
2. Xiamen University, Xiamen 361005, China) |
Abstract: |
In in-depth research on artificial intelligence technology, especially reinforcement learning in the field of energy optimization scheduling, the training efficiency and information utilization rate of the system state expressed as a vector for the learning mode, is low. In response to this problem, this paper proposes an integrated energy system optimization scheduling method that integrates the graph neural network model and reinforcement learning. First, the electricity-heat-gas integrated energy system is modeled as graph structure data, making full use of the topological information of the system. Second, a reinforcement learning model based on graph neural network architecture is proposed, so that it can make full use of graph structure information to achieve a faster training speed and obtain a larger exploration space. Finally, the graph structure information representing the system state is sent to the model for training, and the example simulation verifies the training efficiency and exploration ability of the method. |
Key words: electricity-heat-gas integrated energy system optimal scheduling deep reinforcement learning graph neural network model |