基于CNN-LSTM算法的电力系统气象灾害故障预测方法  

Power system failure prediction under meteorological disasters based on CNN-LSTM neural network

作　　者：赵建伟 林雨场 陈升 李琦 李更丰[2] 张理寅 陆旭 辛正堃 ZHAO Jianwei;LIN Yuchang;CHEN Sheng;LI Qi;LI Gengfeng;ZHANG Liyin;LU Xu;XIN Zhengkun(Xiamen Great Power Geo Information Technology Co.,Ltd.,Xiamen 361012,China;School of Electrical Engineering,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]厦门亿力吉奥信息科技有限公司,福建厦门361012 [2]西安交通大学电气工程学院,陕西西安710049

出　　处：《电工电能新技术》2025年第3期109-118,共10页Advanced Technology of Electrical Engineering and Energy

基　　金：国家电网公司信息通信产业集团能源“双碳”数智化关键技术研发专项(K102200002)。

摘　　要：近年来,日益频发的台风、冰灾、地震、高温等气象灾害事件严重威胁电力系统的安全可靠运行,极端气象灾害下的大规模电网事故导致极高的社会经济损失,因此,对电力系统灾害故障进行精准有效预测具有重要意义。然而,传统方法考虑的故障影响因素类型较为单一,未能同时考虑气象、地理、电网等多种因素对系统故障的影响。同时,考虑极端气象灾害的空间分布和时序演变特性,故障的时空相关性也是预测中的关键因素。因此,本文提出一种基于卷积-长短期记忆神经网络的电力系统气象灾害故障预测方法,建立包含气象、地理、电网多源数据的电力系统故障预测数据集,提出基于卷积神经网络的多源数据融合分析方法,实现故障空间相关性的高效提取;基于长短期记忆算法设计了具有双层网络结构的故障时序预测方法,实现了故障时间相关性的有效刻画,最终形成卷积-长短期记忆神经网络统一框架,提升气象灾害故障预测的准确度。所提方法的有效性和准确性通过台风“米卡拉”、“卢碧”的历史气象数据以及中国东南沿海某区域地理、电网数据进行验证。In recent years,increasingly frequent meteorological disasters such as typhoons,ice disasters,earthquakes,and high temperatures have seriously threatened the safe and reliable operation of the power system.Large-scale power grid accidents under extreme meteorological disasters have resulted in extremely high social and economic losses.Therefore,accurate and effective power system failure prediction method is of great significance.However,the traditional method considers a relatively single type of failure influencing factors and fails to simultaneously consider multiple factors such as meteorology,geography,and power grid.Considering the spatial distribution and temporal evolution characteristics of extreme meteorological disasters,the spatiotemporal correlation of failures is also a key factor in prediction.Therefore,this paper proposes a power system failure prediction method meteorological disasters based on convolution long-short term memory neural network.A power system failure prediction data set is established containing meteorological,geographical,and power grid data.This paper proposes a multi-source data analysis method based on convolutional neural network which can efficiently extract the spatial correlation of failures.A failure sequential prediction method with a double-layer network structure is designed based on the long-short term memory algorithm,which achieves effective characterization of failure temporal correlation.Finally a CNN-LSTM framework is proposed to improve accuracy of failure prediction under meteorological disaster.The effectiveness and accuracy of the proposed method are verified through the historical meteorological data of typhoons Mikala and Lubi as well as the geographical and power grid data of a certain area on the southeastern coast of China.

关 键 词：深度学习 卷积神经网络 长短期记忆神经网络 电力系统故障预测 气象灾害 

分 类 号：TM743[电气工程—电力系统及自动化]