基于多数据源融合的电网故障判别与告警技术研究  

Research on Power Grid Fault Identification and Warning Technology Based on Multiple Data Source Fusion

作　　者：朱轶伦 俞一峰 虞明智 杜晟炜 姚高 许杰 Zhu Yilun;Yu Yifeng;Yu Mingzhi;Du Shengwei;Yao Gao;Xu Jie(Zhejiang Huayun Information Technology Co.,Ltd.,Hangzhou Zhejiang 310007,China;State Grid Zhejiang Electric Power Co.,Ltd.,Hangzhou Zhejiang 310007,China;State Grid Jinhua Electric Power Co.,Ltd.,Jinhua Zhejiang 321000,China)

机构地区：[1]浙江华云信息科技有限公司,浙江杭州310007 [2]国网浙江省电力有限公司,浙江杭州310007 [3]国网金华供电公司,浙江金华321000

出　　处：《电气自动化》2024年第2期32-35,39,共5页Electrical Automation

基　　金：国家电网有限公司科技项目(5500-202219275A-2-0-XG);国网浙江省电力有限公司科技项目(5211DS220003)。

摘　　要：针对国家电网故障判别误差率较高的问题,设计一种基于多数据源融合的电网故障判别与告警方案。利用最大离散小波变换技术和长短期记忆网络算法结合的方法提高电网故障判别与告警能力;利用最大重叠离散小波变换技术具有的扩充冗余自成正交特性对故障类型进行划分;将长短期记忆网络算法由单向进程转为双向网络,避免了反馈传输过程中的网络层无法得到合适的偏导数等梯度消失情况。试验结果表明,通过所提算法进行数据质量核查的准确度高达九成以上,表明所提研究系统对解决提升故障判别准确度的提升具有较强的实用性、优越性。To address the issue of high error rates in fault identification of the state grid of China,a power grid fault identification and alarm scheme based on multi data source fusion was designed.The method combined with maximum overlap discrete wavelet transform(MODWT)technology and long short-term memory(LSTM)network algorithm to improve the power grid fault identification and alarm capabilities;MODWT technology was used with expanded redundancy and self-orthogonal characteristics to classify fault types;transforming the LSTM network algorithm from a unidirectional to a bi-directional network avoided the situation where the network layer can not obtain appropriate partial derivatives and other gradients during feedback transmission.The experimental results show that the accuracy of data quality verification through the proposed algorithm is over 90%,indicating that the research system has strong practicality and superiority in improving the accuracy of fault discrimination.

关 键 词：故障判别 最大重叠离散小波变换技术 长短期记忆网络算法 类型划分 双向网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]