Journal of Arid Meteorology

Previous Articles     Next Articles

Thunderstorm Potential Prediction Based on Back Propagation Neural Network  

CHEN Yongwei1ZHENG Tao1WANG Hankun2WANG Qi1LIANG Yaodan3   

  1. 1. Lightning Protection Center of Gansu Province,Lanzhou 730020,China;
    2. Inner Mongolia Linhtning Early Warning Protection Center,Huhehaote 010051,China;
    3. Guigang Meteorological Bureau of Guangxi,Guigang 537100,China
  • Received:2013-03-24 Revised:2013-06-06 Online:2013-09-30 Published:2013-09-30

基于BP神经网络模型的雷电潜势预报

陈勇伟郑 涛王汉王 琦梁耀丹   

  1. 1. 甘肃省防雷中心,甘肃 兰州 730020;
    2. 内蒙古雷电预警防护中心,内蒙古 呼和浩特 010051;
    3. 广西贵港市气象局,广西 贵港 573100
  • 作者简介:陈勇伟(1987 - ),男,助理工程师,主要从事防雷风险评估、防雷图纸审核与防雷设计工作. E - mail:chenyongweiaw@163. com

Abstract:

In order to use neural networks to solve common nonlinear problem in lightning potential trend prediction,the correlation co-
efficients were calculated between forty - six convective parameters and thunderstorms occurring from June to August in 2008 in Nan-
jing. In these convective parameters,seven convective factors among them had better relationship with thunderstorms occurrence,in-
cluding TT,SI,SWEAT,Tlfc,CIN,DCI and PW indexes,then these seven convective parameters were selected as the input factors
of the neural network model which contained seven input layers,twelve hidden layers and one output layer. On the basis of back propa-
gation neural network model built by the data of 2008,the thunderstorm potential trend from June to August in 2009 in Nanjing were
predicted including the thunderstorm days and non - thunderstorm days. According to the score standard,the POD,FAR,CSI,PDFD
and FOM of the model were 74. 5%,9. 5%,74. 5% 2. 9% and 19. 1%,respectively,which indicated that this back propagation neu-
ral network model had better forecast accuracy and its performance was steady,it can be well applied in thunderstorm potential trend
prediction.

 

Key words:  convective parameters, factor combination, thunderstorm potential trend prediction, back propagation neural network model

摘要:

为了使用神经网络较好地解决在雷电潜势预报中常见的非线性问题,本文通过计算南京地区2008 年 6 ~8 月 46 个对流参数与雷电发生的相关系数,选取了与雷电发生关系较好的 TT、SI、CIN 等7 个对流参数作为 BP 神经网络的输入因子。利用 2008 年的资料所建立的 BP 神经网络模型,预报了南京地区 2009 年 6 ~8 月的雷暴活动潜势,结合实际雷暴发生情况,得到此模型的 POD 为 80. 9%,FAR 为 9. 5%,CSI 为 74. 5%,PDFD 为 2. 9%,FOM 为 19.1%。表明该 BP 模型预报准确率较高,性能稳定,有较好的推广价值。

关键词: 对流参数, 因子组合, 雷电潜势预报, BP神经网络

CLC Number: