利用粒子群算法改进电离层TEC预报模型  

Using PSO algorithm to improve the ionosphere TEC forecast model

作　　者：杨淑芬 曾聪 唐钰涵 YANG Shufen;ZENG Cong;TANG Yuhan(Second Highway Survey and Design Research Institute, China Communications Construction, Wuhan 430056, China)

机构地区：[1]中交第二公路勘察设计研究院有限公司,武汉430056

出　　处：《测绘工程》2022年第2期24-30,共7页Engineering of Surveying and Mapping

基　　金：国家自然科学基金资助项目(41761089)。

摘　　要：针对电离层TEC预报神经网络模型参数选择复杂度高的问题,引入粒子群算法优化改进LSTM神经网络模型参数,以提高其预报精度。利用IGS中心提供的低、中、高纬度电离层TEC数据,根据太阳活动选取两个时段进行预报,将预报结果与IGS中心提供的TEC值进行对比分析。实验结果表明,PSO-LSTM模型预报效果最优,太阳活动平静期预报均方根误差为0.81 TECu,平均相对精度为91.72%;太阳活动剧烈期预报均方根误差为1.25 TECu,平均相对精度为80.98%,通过对比分析表明改进模型在预报精度和稳定性方面相比传统模型均有提升。Aimed at the problem of high parameter selection complexity of the ionospheric TEC prediction neural network model,a particle swarm optimization algorithm is introduced to improve the parameters of LSTM neural network model to improve its prediction accuracy.Based on the ionospheric TEC data of low,middle and high latitude provided by IGS Center,two periods of solar activity are selected for the forecast,and the result is compared with the TEC values provided by IGS Center.The experimental result shows that the PSO-LSTM model has the best prediction effect,and the root mean square error of solar quiet period prediction is 0.81 TECu,with which the average relative accuracy is 91.72%.The root-mean-square error of solar activity prediction is 1.25 TECu,and the average relative accuracy is 80.98%.Comparative analysis shows that the improved model has better prediction accuracy and stability than the traditional model.

关 键 词：粒子群算法 LSTM神经网络 总电子含量 预报精度 

分 类 号：P228[天文地球—大地测量学与测量工程]