基于深度学习的雷电活动预测方法及其输电线路防雷应用  

Lightning Activity Prediction Method Based on Deep Learning and its Application in Transmission Line Lightning Protection

作　　者：张永刚 谷山强 李健[1,2,3] 吴大伟 王宇[1,2,3] ZHANG Yonggang;GU Shanqiang;LI Jian;WU Dawei;WANG Yu(NARI Group Corporation(Stage Grid Electric Power Research Institute),Nanjiang 211106,China;Wuhan NARI Limited Company,State Grid Electric Power Research Institute,Wuhan 430074,China;Hubei Key Laboratory of Power Grid Lightning Risk Prevention,Wuhan 430074,China)

机构地区：[1]南瑞集团有限公司(国网电力科学研究院有限公司),南京211106 [2]国网电力科学研究院武汉南瑞有限责任公司,武汉430074 [3]电网雷击风险预防湖北省重点实验室,武汉430074

出　　处：《电瓷避雷器》2024年第4期18-28,共11页Insulators and Surge Arresters

基　　金：国家自然科学基金项目(编号:52007037)。

摘　　要：为提升输电线路雷击灾害主动防御能力,提出了一种基于深度学习的雷电活动预测方法,提前时间为72 h,时间和空间精度分别为3 h和5 km。基于统一的时空网格完成了预报区域雷电数据的归一化,通过卡方统一性检验提取了与雷电活动强关联的气象参量;建立了雷电发生概率预测深度神经网络模型,采用贝叶斯算法优化了模型超参数组合;建立了落雷次数与雷电流强度的分类预测卷积神经网络模型。算例验证表明,雷电发生概率预测的平均命中率和虚警率分别为69.10%和71.18%,落雷次数和雷电流强度预测命中率的平均值分别为39.03%和37.94%,提前预测超高压线路雷击跳闸的准确率为87.5%,平均距离误差为4.01 km。本方法可用于开展基于预报信息的输电线路雷击故障主动防护,对降低雷击灾害损失、提升线路防雷水平具有十分重要的意义。In order to improve the active protection ability of transmission lines against lightning disaster,a lightning activity prediction method based on Deep Learning was proposed,the lead time is 72 h,and the time and spatial accuracy are 3 h and 5 km,respectively.Based on the unified spatio-temporal grid,the lightning data in the forecast area were normalized,and the meteorological parameters strongly correlated with lightning activity were extracted by Chi-square unity test.A deep neural network model for lightning occurrence probability prediction was established,and the hyper-parameters combination of the model was optimized by Bayesian algorithm.A Convolutional Neural Network model is established to predict the number of lightning falls and the intensity of lightning current.The calculation results show that the probability of detection and false alarm rate of the prediction model are 69.10%and 71.18%,and the average score of the prediction model for the number of lightning falls and lightning current intensity is 39.03%and 37.94%.The accuracy of the prediction for the lightning trip of Ultra high voltage lines is 87.5%,and the mean distance error is 4.01km.This method can be used to carry out lightning fault active protection of transmission lines based on forecast information,which is of great significance to reduce lightning disaster loss and improve lightning protection level of transmission network.

关 键 词：输电线路 雷电预报 中尺度气象模式 深度学习 卡方检验 贝叶斯优化 

分 类 号：TM863[电气工程—高电压与绝缘技术] TP18[自动化与计算机技术—控制理论与控制工程]