利用PSO算法改进的ENN预报中国区域电离层TEC
Ionospheric TEC Prediction in China Based on ENN Improved by PSO Algorithm
-
摘要: 精确的电离层总电子含量（total electron content,TEC）预报,对提高全球导航卫星系统（global navigation satellite system,GNSS）的导航定位精度具有重要意义。提出了一种采用基于粒子群优化（particle swarm optimization,PSO）算法的Elman神经网络（Elman neural network,ENN）预报电离层TEC的方法。首先,利用中国大陆构造环境监测网络（crustal movement observation network of China,CMONOC）2018年的GNSS观测数据构建中国区域电离层模型,获取区域电离层TEC格网（regional ionospheric map,RIM）。其次,根据RIM TEC数据建立PSO-ENN模型进行预报研究。实验结果表明,磁静日和磁暴日下ENN模型和PSO-ENN模型的均方根误差（root mean square error,RMSE）均在训练集时间尺度为21 d时最低,使用PSO-ENN模型滑动预报5 d的RMSE相较ENN模型分别降低27.6%和20.5%。同时,采用欧洲定轨中心（center for orbit determination in Europe,CODE）发布的全球电离层TEC格网（global ionospheric map,GIM）产品建立PSO-ENN模型对中国区域电离层TEC进行预报时,磁静日和磁暴日的平均RMSE均小于CODE发布的1 d预报产品（CODE’s 1-day predicted GIM,C1PG）、BP神经网络（back propagation neural network,BPNN）模型和ENN模型,说明PSO-ENN模型的预报效果优于C1PG、BPNN模型和ENN模型。利用RIM数据对2018年8月份TEC进行30 d滑动预报时,PSO-ENN模型的平均RMSE相较BPNN模型和ENN模型分别降低24.8%和14.3%,说明PSO-ENN模型具有较高的稳定性。Abstract:Objectives Accurate prediction of ionospheric total electron content (TEC) is of great research significance to improve the accuracy of global navigation satellite system (GNSS) navigation and positioning.Methods We propose an ionospheric TEC prediction model based on particle swarm optimization (PSO) algorithm and Elman neural network (ENN). First, high accurate regional ionospheric map (RIM) in China is set up by using the GNSS observation data from the crustal movement observation network of China (CMONOC) in 2018. Second, the RIM TEC data are used for forecasting research.Results During the quiet period and disturbance period, the best time scale for training set is 21 d. For the 5 d sliding prediction, the root mean square error (RMSE) of the PSO-ENN model during the quiet period and disturbance period are decreased by 27.6% and 20.5%, respectively. When using the global ionospheric map (GIM) published by center for orbit determination in Europe (CODE) in China region for different grid points prediction, the average RMSE of the PSO-ENN model is lower than the CODE 1-day predicted GIM (C1PG) by back propagation neural network (BPNN) model and ENN model. For the 30-day sliding prediction in August 2018 based on RIM TEC in China region, the average RMSE of the PSO-ENN model are 24.8% and 14.3% lower than that of BPNN and ENN models, respectively.Conclusions The proposed model has better prediction accuracy and stability both in the quiet period and disturbance period.
下载:
