基于机器学习和单站地面气象要素的雷电临近预警方法  

Warning Method of Lightning Nowcasting Based on Machine Learning and Single Station Ground Meteorological Elements

作　　者：赵生昊 覃彬全 杜乐 ZHAO Shenghao;QIN Binquan;DU Le(Chongqing Weather Safety Technology Center,Chongqing 401120;Chongqing Bureau of Geology and Minerals Exploration,Chongqing 401121)

机构地区：[1]重庆市气象安全技术中心,重庆401120 [2]重庆市地质矿产勘查开发局,重庆401121

出　　处：《气象科技》2022年第1期121-128,共8页Meteorological Science and Technology

基　　金：重庆市气象部门业务技术攻关项目(YWJSGG-202146);重庆“两江之星”气象英才计划资助。

摘　　要：为减少雷电灾害造成的人身伤亡和经济损失,提出了一种基于机器学习和单站地面气象要素的雷电临近预警方法。在重庆市选择了8个自动气象站,使用温、压、湿、风4种单站地面气象要素与ADTD地闪定位资料,通过对特征工程、重采样、交叉验证等机器学习技术的组合应用,构建了基于ADASYN-ET模型的雷电临近预警方法,能够对气象站周边20 km范围进行提前期0~30 min的预警。验证结果表明:该预警方法适用于全部8个站点,在0~10 min、10~20 min、20~30 min预警提前期,调和平均F;得分分别为0.60、0.59、0.60,与其他一些预警方法、系统对比实现方式灵活,预警效果良好,能够为防雷减灾工作提供参考。In order to reduce personal casualties and economic losses caused by lightning disasters,a lightning nowcasting method based on the ADASYN-ET model is proposed.Using four single-station ground meteorological elements of temperature,pressure,humidity and wind combined with ADTD ground-flash location information,a lightning nowcasting method based on the ADASYN-ET model is constructed through the combined application of machine learning techniques such as feature engineering,resampling,and cross-validation,which can provide 0 to 30 min advance warning for the 20 km range of weather stations.The validation results show that the warning method applies to all 8 test sites,and the average F;Score is 0.60,0.59 and 0.60,0 to 10 min,10 to 20 min and 20 to 30 min in advance.The method is flexible compared to other recent warning methods and systems for implementation and can provide references for the work of lightning prevention and disaster reduction.

关 键 词：雷电临近预警 气象要素 闪电定位系统 机器学习 

分 类 号：P429[天文地球—大气科学及气象学]