摘要
传统配电网的单一目标调度在负荷波动加剧、分布式能源接入及极端事件频发时存在明显局限,因此提出基于深度学习的动态重构与多目标优化方案。采用长短期记忆网络(long short-term memory,LSTM)-图卷积网络(graph convolutional network,GCN)混合模型感知配电网时空状态,构建经济性、可靠性、低碳性的多目标模型,融合深度强化学习(deep reinforcement learning,DRL)与非支配排序遗传(NSGA-Ⅱ)算法实现策略搜索与Pareto解集生成。试验表明:该方法决策时间短、网损低、可再生能源消纳率高,极端工况下鲁棒性强,为配电网智能化运行提供创新方案。
In response to the limitations of traditional distribution network scheduling in load fluctuations,distributed energy access,and extreme events,a dynamic reconstruction and multi-objective optimization scheme based on deep learning was proposed.The spatiotemporal state of the distribution network was perceived by long short-term memory(LSTM)and graph convolutional network(GCN)model so as to construct a multi-objective model integrating deep reinforcement learning(DRL)and improved non-dominated sorting genetic algorithmⅡ(NSGA-Ⅱ)which included economy,reliability,and low carbon to achieve policy search and Pareto solution set generation.The experiment shows that this method takes less decision-making time,lower network loss,higher renewable energy consumption rate,and stronger robustness under extreme working conditions,which provides innovative solutions for intelligent operation of distribution networks.
作者
唐爽
TANG Shuang(State Grid Shaanxi Electric Power Co.,Ltd.,Zhenba County Power Supply Branch,Hanzhong 723600,China)
出处
《技术与市场》
2025年第12期69-73,共5页
Technology and Market
关键词
深度学习
配电网自动化
动态重构
多目标优化调度
deep learning
distribution network automation
dynamic reconstruction
multi objective optimization scheduling
