基于卷积神经网络的典型雷暴云下地面电场识别研究

Recognition of Ground Electric Field Under Typical Thunderstorm Clouds Based on Convolutional Neural Network

雷暴云产生的闪电危害巨大,雷暴发生时其地面电场的特征可应用于关键系统的雷电预警和雷电防护。笔者依据雷暴云偶极子模型,研究典型雷暴云下地面电场的特征,提出基于卷积神经网络(Convolutional Neural Network,CNN)的雷暴云地面电场识别方法。该方法考虑不同天气情况下地面电场的特征,对大量样本数据进行训练和识别,证明了卷积神经网络用于识别雷暴云下地面电场的可行性,划分了基于卷积神经网络识别结果的雷电预警等级。结果表明:所提方法区分各类别电场样本的整体准确率高于93%,通过对典型雷暴云下地面电场特征的学习,实现了雷暴云电荷中心的地理区域识别。所提方法可为局部区域的雷暴云大气电场识别和雷电预警提供思路。

Lightning originated from thunderstorm clouds is harmful,and the associated ground electric field can offer valuable guidance for lightning warning and lightning protection.The characteristics of the ground electric filed under typical thunderstorm clouds are studied on a basis of the classic dipole thundercloud model.The author proposed a method for recognizing the ground electric field of typical thunderstorm clouds based on the Convolutional Neural Network(CNN).A number of data samples of ground electric field under different synoptic conditions are trained and recognized by a CNN model in the proposed method,proving the feasibility of CNN method in recognizing different samples.The classification of lightning warning levels is suggested based on the recognition results.Results show that,the overall accuracy rate of the proposed method is higher than 93%in distinguishing the different types of electric field samples.The geographical area of the charge center in thunderstorm clouds can be recognized through the characteristics learning of the ground electric field.The proposed method can serve to the recognition of thunderstorm clouds and lighting warning in local areas.