首页|期刊导航|北京测绘|面向电离层TEC预报的CEEMDAN-GRNN-BiLSTM组合模型

面向电离层TEC预报的CEEMDAN-GRNN-BiLSTM组合模型OA

A combined CEEMDAN-GRNN-BiLSTM model for ionospheric TEC forecasting

中文摘要英文摘要

为提高电离层总电子含量(TEC)的预报精度及其在地磁扰动条件下的鲁棒性,本文提出一种面向电离层TEC预报的自适应噪声完备集合经验模态分解-广义回归神经网络-双向长短期记忆网络(CEEMDAN-GRNN-BiLSTM)组合模型.采用CEEMDAN方法对原始电离层TEC序列进行处理,得到一系列本征模态函数(IMF);基于样本熵算法评估各IMF分量的复杂度,完成分量的自适应重组;对重组高频分量选用GRNN建立预测模型,低频分量选用BiLSTM进行建模与预报;综合各分量的预报结果,重构出完整的电离层TEC预报值.为验证模型的有效性,选取不同地磁活动水平与不同纬度的实测TEC数据开展实验.结果表明:在磁平静期高纬区域,本文模型的均方根误差(RMSE)为0.41 TECu,较单一模型分别降低了0.11 TECu与0.05 TECu;在磁暴期高纬区域,模型的RMSE为0.62 TECu,较单一模型分别降低了0.28 TECu与0.15 TECu,体现出本文模型在不同空间天气条件下的适应性与可靠性.

To enhance the forecasting accuracy of ionospheric total electron content(TEC)and the robustness of forecasting outcomes under geomagnetic disturbance conditions,this paper proposed a combined model integrating complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),generalized regression neural network(GRNN),and bidi-rectional long short-term memory network(BiLSTM)for ionospheric TEC forecasting.The raw ionospheric TEC time series was first decomposed via the CEEMDAN method to obtain a series of intrinsic mode functions(IMFs).The complexity of each IMF component was then evaluated based on the sample entropy algorithm,followed by adaptive recombination of the components.For the recombined high-frequency components,a GRNN-based forecasting model was constructed,while the low-frequency components were modeled and forecasted using the BiLSTM network.Finally,the forecast results of all com-ponents were integrated to reconstruct the complete ionospheric TEC forecast values.To validate the effectiveness of the pro-posed model,measured TEC data from different geomagnetic activity levels and at different latitudes were selected for the experiments.The results show that,during geomagnetically quiet periods,the model achieves a root mean square error(RMSE)of 0.41 TECu,which is 0.11 TECu lower than that of a single-model baseline and 0.05 TECu lower than the other.During geomagnetic storm periods,the model's RMSE is 0.62 TECu,representing reductions of 0.28 TECu and 0.15 TECu relative to the two single-model baselines,respectively.These improvements demonstrate the model's adaptabil-ity and reliability under different space-weather conditions.

宋晓蛟;鲜高强;蒋国政

陕西地矿第二综合物探大队有限公司,陕西 西安 710016广州南方测绘科技股份有限公司西安分公司,陕西 西安 710100广州南方测绘科技股份有限公司西安分公司,陕西 西安 710100

天文与地球科学

自适应噪声完备集合经验模态分解(CEEMDAN)电离层总电子含量(TEC)预报广义回归神经网络(GRNN)双向长短期记忆网络(BiLSTM)

complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)ionospheric total electron content(TEC)forecastinggeneralized regression neural network(GRNN)bidirectional long short-term memory network(BiLSTM)

《北京测绘》 2026 (8)

1089-1096,8

陕西省自然科学基础研究计划项目(2021JQ562)

10.19580/j.cnki.1007-3000.2025090018

评论