一种雷达回波外推短临预报方法仿真  

Simulation of a short-term and imminent prediction method based on radar echo extrapolation

作　　者：郭艳萍[1] 高云 周建慧 彭炜[1] GUO Yan-ping;GAO Yun;ZHOU Jian-hui;PENG Wei(School of Computer and Network Engineering,hanxi Datong University Datong Shanxi 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》2022年第2期1-4,16,共5页Computer Simulation

基　　金：山西省软科学研究计划项目[2019041023-5]。

摘　　要：雷达回波外推结果是短临预报的基础,为获取准确短临预报内容,并提升预报内容的时效性,研究基于循环动态卷积的雷达回波外推短临预报方法。在动态卷积神经网络的基础上,参考循环神经网络结构特性,构建循环动态卷积神经网络,在卷积层内引入循环结构,生成外推雷达回波图像同过去一段时间内输入雷达回波图像序列间的相关性。在循环动态子网络内输入采集到的雷达回波图像,通过循环动态子网络处理获取两个概率向量,以其作为概率外推层的卷积核,并与输入图像序列内最后一幅图像实施卷积处理,获取雷达回波图像外推结果。依据外推结果同短临预报内容间的相关性,生成短临预报内容。仿真结果显示,上述方法外推结果与实际观测获取的雷达回波图像一致度更高,且时效更长。For improving the accuracy and timeliness of short-term and imminent prediction, this paper studies the short-term and imminent prediction method of radar echo extrapolation based on cyclic dynamic convolution. Based on the structural characteristics of dynamic convolution neural network and cyclic neural network, a cyclic dynamic convolution neural network was built. The convolution layer was introduced into the cyclic structure to generate the correlation between the extrapolated radar echo image and the input radar echo image sequence in the past period. The collected radar echo image was input into the cyclic dynamic subnetwork. According to the cyclic dynamic subnetwork, two probability(convolution kernel of probability extrapolation layer) vectors were obtained and convoluted with the last image in the input image sequence, completing the extrapolation result of the radar echo image. Finally, the short-term and imminent forecast content was generated by the correlation between the extrapolation results and the short-term and imminent forecast content. The simulation results show that this method has the content of short-term and imminent prediction with long time effect, and the extrapolation results are closer to the actual observated results.

关 键 词：循环动态卷积 雷达回波外推 短临预报 概率向量 卷积处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]