基于时空图卷积网络和机器学习的跨地区电价异常识别

陈思宇; 黄绪荣; 刘颖; 陈璐; 覃恺; 谢佩

陈思宇,黄绪荣,刘颖,陈璐,覃恺,谢佩.基于时空图卷积网络和机器学习的跨地区电价异常识别[J].电测与仪表,2026,63(9):87-95.
Chen Siyu,Huang Xurong,Liu Ying,Chen Lu,Qin Kai,Xie Pei.Cross regional electricity price anomaly recognition based on spatiotemporal graph convolutional network and machine learning[J].Electrical Measurement & Instrumentation,2026,63(9):87-95.

基于时空图卷积网络和机器学习的跨地区电价异常识别

Cross regional electricity price anomaly recognition based on spatiotemporal graph convolutional network and machine learning

DOI：10.19753/j.issn1001-1390.2026.09.009

中文关键词: 时空图卷积网络机器学习支持向量机电价异常识别差分进化

英文关键词:spatiotemporal graph convolutional network, machine learning, support vector machine, identification of abnormal electricity price, differential evolution

基金项目:广西电网公司科技项目（项目编号：044600KC23040002）

作者中文名	作者英文名	单位
陈思宇	Chen Siyu	广西电网有限责任公司
黄绪荣	Huang Xurong	广西电网有限责任公司
刘颖	Liu Ying	广西电网有限责任公司贵港供电局
陈璐	Chen Lu	广西电网有限责任公司南宁供电局
覃恺	Qin Kai	广西电网有限责任公司
谢佩	Xie Pei	广西电网有限责任公司

摘要点击次数: 68

中文摘要:

传统跨地区电价异常识别方法受限于数据处理和分析的浅层次特性，难以深入挖掘电价数据中的深层次特征和潜在模式，导致电价异常识别精度较低。为此，提出一种基于时空图卷积网络和机器学习的跨地区电价异常识别方法。将电力系统网络拓扑看作图结构，通过傅里叶变换得到电价数据时空图矩阵，融合空间图卷积层和时间门控卷积层创建时空图卷积网络，通过网络学习输出跨地区电价异常时空特征；组建基于支持向量机的跨地区电价异常识别模型，运用差分进化的变异、交叉和选择环节寻优模型参数，训练电价样本数据集得到异常识别结果。实验结果表明，所提方法电价异常识别精度高，识别效率高，可为电力企业能源管理和数据分析提供参考。

英文摘要:

Traditional cross regional electricity price anomaly identification methods are limited by the shallow characteristics of data processing and analysis, making it difficult to deeply explore the deep features and potential patterns in electricity price data, resulting in low accuracy in electricity price anomaly identification. To this end, a cross regional electricity price anomaly recognition method based on spatiotemporal graph convolutional network and machine learning is proposed. Viewing the topology of the power system network as a graph structure, the spatiotemporal graph matrix of electricity price data is obtained through Fourier transform, a spatiotemporal graph convolutional network is created by integrating spatial graph convolutional layers and time gated convolutional layers, and the spatiotemporal characteristics of cross regional electricity price anomalies are output through network learning; a cross regional electricity price anomaly recognition model is built based on support vector machines, mutation, crossover, and selection processes of differential evolution are used to optimize model parameters, and an electricity price sample dataset is trained to obtain anomaly recognition results. The experimental results show that the proposed method has high accuracy and fast recognition efficiency in identifying abnormal electricity prices, and can provide reference for energy management and data analysis in power enterprises.

查看全文查看/发表评论下载PDF阅读器

关闭