###
DOI:
本文二维码信息
基于卷积神经网络(CNN)的雷电临近预警模型
张烨方
(福建省气象科学研究所)
Lightning proximity early warning model based on convolutionalneural network (CNN)
Zhang Yefang
(Fujian Institute of Meteorology and Science,Fujian Fuzhou)
摘要
相似文献
本文已被:浏览 3次   下载 18
投稿时间:2019-11-27    修订日期:2021-01-24
中文摘要: 文章从研究人工智能雷电临近预警模型的目的出发,以卷积神经网络模型为基础,结合多个时间序列的雷达(MCR、VIL、ET)与闪电数据,对雷电临近预报方法进行基于卷积神经网络结构的应用,以福建省2017、2018年雷达、闪电数据为样本完成了模型的训练与预测研究。训练结果显示,15-30mins模型训练样本测试集准确率79.85%;选取福建省2019年20天雷电过程验证分析表明,15-30mins模型对动力抬升型雷电过程预警TS评分0.716,夏季局地热雷暴预警TS评分0.694,与常规采用雷达、闪电阈值控制的雷电预警算法相比准确率有一定的提高,具有一定的实践意义。
Abstract:Based on the purpose of studying the near-warning model of artificial intelligence lightning, based on the convolutional neural network model, combining radar (MCR, VIL, ET) and lightning data of multiple time series, the application of the lightning proximity prediction method based on the structure of convolutional neural network is carried out, and the radar structure based on the lightning in Fujian Province in 2017 and 2018, Lightning data completes the training and prediction of models for the sample. The training results show that the test set accuracy of 15-30mins model training sample is 79.85 percent, and The analysis of the 20-day lightning process verification in Fujian Province of 2019 shows that the TS score of the 15-30mins model for the early warning of the power-lift lightning process is 0.716, and the TS score of the geothermal thunderstorm warning in summer bureau is 0.694, with the conventional radar. The lightning early warning algorithm controlled by lightning threshold has a certain improvement in accuracy and has certain practical significance.
文章编号:201911270417     中图分类号:    文献标志码:
基金项目:福建省科技厅社会发展引导性(重点)项目(2019Y0063)资助
引用文本:
张烨方,0.Lightning proximity early warning model based on convolutionalneural network (CNN)[J].Meteor Mon,():-.
Zhang Yefang,0.Lightning proximity early warning model based on convolutionalneural network (CNN)[J].Meteor Mon,():-.