基于卷积神经网络的天气雷达回波外推模型  

Weather Radar Echo Extrapolation Model Based on Convolution Neural Network

作　　者：高云 郭艳萍[1] 张叶娥 GAO Yun;GUO Yan-ping;ZHANG Ye-e(School of Computer and Network Engineering,Shanxi Datong University,Shanxi Datong 037009,China)

机构地区：[1]山西大同大学计算机与网络工程学院,山西大同037009

出　　处：《计算机仿真》2021年第12期272-275,285,共5页Computer Simulation

基　　金：山西省软科学研究计划项目(2019041023-5)。

摘　　要：常用天气雷达回波外推方法建立的天气雷达回波外推模型,设置编码器和预测器长度和步长偏小,导致模型外推预测天气精度低,于是提出基于卷积神经网络的天气雷达回波外推模型。转换雷达图像数据集坐标,计算天气变化粒子的经纬度值,修正天气变化粒子距离地面高度,预处理天气雷达图像数据;将卷积神经网络分为正向传播和反向传播两种,分别训练卷积神经网络数据传播过程;采用卷积层和采样层交替布置的方式,设置编码器和预测器长度和步长,建立天气雷达回波外推模型预测天气。实验结果表明:对比三组天气雷达回波外推模型,所设计模型具有较高的外推预测精度。Currently, the weather radar echo extrapolation model has low prediction accuracy owing to the small length and step size of the encoder and predictor. Therefore, this paper reports the weather radar echo extrapolation model based on convolution neural network. The data set coordinates of the radar image were transformed. The longitude and latitude values of weather changing particles were calculated. The height of weather change particles from the ground was corrected, and the image data of weather radar were preprocessed. The data propagation process of convolution neural network was trained by forward and backward propagation of convolution neural network. According to the alternate arrangement of convolution layer and sampling layer, the length and step size of encoder and predictor were set, and the weather prediction model of weather radar echo extrapolation was established. The experimental results show that the prediction accuracy of this model is higher than that of the traditional model.

关 键 词：卷积神经网络 天气雷达 回波 外推模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]