融合电离层参数相似特征的f_(0)F_(2)参数深度学习预测方法

Deep Learning Prediction Method for f_(0)F_(2)Parameters Based on the Ionospheric Parameter Similarity Features

电离层临界频率f_(0)F_(2)参数是重要的电离层参数之一,开展f_(0)F_(2)参数预测具有重要的研究意义和应用价值.提出了一种融合f_(0)F_(2)参数变化特性的深度学习预测方法,采用双向长短时记忆神经网络(BiLSTM)和电离层参数相似特征相结合的模型实现电离层临界频率f_(0)F_(2)参数提前24 h预测.结果表明,BiLSTM结合电离层参数相似特征模型预测f_(0)F_(2)参数的平均相对误差在8%~10%.对不同纬度的探测站的f_(0)F_(2)参数预测结果表明,随着纬度降低,预测难度和误差都会增大,预测精度降低.对地磁暴期间的f_(0)F_(2)参数预测结果分析发现,地磁暴期间的预测效果会受到一定程度的干扰,预测误差增大.在地磁暴期间,相比长短时记忆神经网络(LSTM)模型和BiL-STM模型,BiLSTM结合电离层参数相似特征模型对于f_(0)F_(2)参数的预测效果更优.

The ionosphere is an important component of the solar terrestrial space environment,and the critical frequency f_(0)F_(2) parameter is one of the most important and complex ionospheric parameters.The changes in f_(0)F_(2) parameters will have a certain degree of impact on communication,navigation,radar and other technologies,so predicting f_(0)F_(2) parameters has important research significance and application value.This article proposes a deep learning prediction method that integrates the characteristics of f_(0)F_(2) parameter changes.A model combining bidirectional long short-term memory neural network BiL-STM network and ionospheric parameter similarity features is used to predict the ionospheric critical frequency f_(0)F_(2) parameter 24 hours in advance.The results show that the average relative error of BiLSTM combined with ionospheric parameter similarity model in predicting f_(0)F_(2) parameters is about 8%~10%.Compared with the Long Short-Term Memory(LSTM)model,the average relative error has decreased by about 6%to 7%,while compared with the BiLSTM model,the average relative error has decreased by about 4%to 5%.The prediction results of the f_(0)F_(2) parameter for different latitude detection stations show that as the latitude decreases,the difficulty of predicting the f_(0)F_(2) parameter increases,the prediction errors of the three models increase,and the prediction accuracy decreases.The analysis of the prediction results of f_(0)F_(2) parameters during geomagnetic storms shows that the predictive performance of the three models will be affected to varying degrees during the occurrence of geomagnetic storms,and the prediction error will increase.Compared to the calm period,the average relative error of the three models has increased by about 1%to 4%.During geomagnetic storms,compared with LSTM and BiLSTM models,BiLSTM combined with ionospheric parameter similarity feature models has better predictive performance for f_(0)F_(2) parameters and better predictive performance.This method can also be applied to the prediction research of other ionospheric parameters such as Total Electron Content(TEC),hmF2,etc.,and has a very broad application prospect.