基于LSTM和随机森林的避雷器故障预警算法  

Fault warning algorithm of arrester based on LSTM and Random Forest

作　　者：刘志伟[1] 宁克[1] 刘星廷 侯滨[1] 王海旗 LIU Zhiwei;NING Ke;LIU Xingting;HOU Bin;WANG Haiqi(Materials Branch,State Grid Shanxi Electric Power Co.,Ltd.,Taiyuan 030021,China;Electric Power Science Research Institute,State Grid Shanxi Electric Power Co.,Ltd.,Taiyuan 030021,China)

机构地区：[1]国网山西省电力公司物资分公司,山西太原030021 [2]国网山西省电力公司电力科学研究院,山西太原030021

出　　处：《电子设计工程》2024年第22期137-141,共5页Electronic Design Engineering

基　　金：国网山西省电力公司科技项目(B9QD-300009601-00001)。

摘　　要：针对传统实验方法无法准确预测避雷器状态及当前在线监测方法易受环境因素干扰的问题,提出了一种基于长短期记忆(LSTM)网络与随机森林(RF)算法的避雷器故障预警模型。该模型利用LSTM算法通过避雷器的关键特征量,对其未来状态做出预测,并将LSTM的预测数据输入到预训练好的改进随机森林模型进行故障类型分析,实现提前告警。多组对比实验结果表明,所提方法对避雷器故障预测的平均绝对百分比误差(MAPE)范围为4.16%~5.62%,均方根误差(RMSE)范围为0.136~0.154,而针对避雷器故障分类的总体准确率为92.6%,有效实现了避雷器的状态预测和故障分类,可以为工程应用提供更为精准的决策依据。In order to solve the problem that traditional test methods cannot accurately predict the status of lightning arrester and current online monitoring methods are vulnerable to environmental factors,a lightning arrester fault early warning model based on Long Short⁃Term Memory(LSTM)network and Random Forest(RF)algorithm is proposed.This model uses LSTM to predict the future state of the arrester through its key characteristics,and inputs LSTM prediction data into the pre trained improved random forest model for fault type analysis to achieve early warning.Multiple sets of comparative experimental results show that the Mean Absolute Percentage Error(MAPE)of the proposed method for lightning arrester fault prediction ranges from 4.16%to 5.62%,the Root Mean Square Error(RMSE)ranges from 0.136 to 0.154,and the overall accuracy rate for lightning arrester fault classification is 92.6%,which effectively realizes the status prediction and fault classification of lightning arrester,and can provide more accurate decision⁃making basis for engineering applications.

关 键 词：避雷器 在线监测 故障预警 长短期记忆 随机森林算法 

分 类 号：TN911.72[电子电信—通信与信息系统] TM862[电子电信—信息与通信工程]