A Prediction Model of Ionospheric foF2 Based on Extreme Learning Machine. Issue 10 (27th October 2018)

• Record Type: Journal Article
• Authors: Bai, Hongmei; Fu, Haipeng; Wang, Jian; Ma, Kaixue; Wu, Taosuo; Ma, Jianguo
• Is Part Of: Radio science. Volume 53:Issue 10(2018)
• Page Start: 1292
• Page End: 1301
• Publication Date: 2018-10-27
• Subjects: ionosphere -- foF2 -- neural network -- extreme learning machine -- Darwin; Radio meteorology -- Periodicals; Radio wave propagation -- Periodicals; 621.38405
• DOI: 10.1029/2018RS006622
• Languages: English
• ISSNs: 0048-6604

Abstract

Abstract: The highly nonlinear variation of the ionospheric F 2 layer critical frequency ( f o F 2 ) greatly limits the efficiency of communications, radar, and navigation systems that employ high‐frequency radio waves. This paper proposes an effective method to predict the f o F 2 using the extreme learning machine (ELM). Compared with the previous neural network model based on feedforward algorithm, the ELM model offers the advantages of faster training speed and less manual intervention. The ELM model is trained with the daily hourly values of f o F 2 at Darwin (12.4°S, 131.5°E) in Australia. The training data are selected from 1995 to 2012, except 1997 and 2000, which includes all periods of quiet and disturbed geomagnetic conditions. The f o F 2 data to verify model performance are selected in 1997, 2000, and 2013, which are low, high, and moderate solar activity years, respectively. The prediction results have shown that the proposed ELM model can achieve faster training process while maintaining the similar accuracy compared with BPNN. In addition, the proposed ELM model is compared with the International Reference Ionosphere model prediction. The ELM model predicts the f o F 2 values more accurately than the International Reference Ionosphere model in low (1997), moderate (2013), and high (2000) solar activity years, as clearly seen on the yearly root‐mean‐square error. As far as the author's knowledge, this is the first time that the ELM model is applied to predict f o F 2 . Plain Language Summary: The spatiotemporal variability of the ionospheric F 2 layer critical frequency ( f o F 2 ) affects the efficiency of communications, radar, and navigation systems that employ high‐frequency radio waves. Therefore, many researchers have established different kinds of models to predict the changes in f o F 2 based on historical data. However, these studies only focus on how to improve the prediction accuracy of the model. In fact, for a predictive model, it is also crucial to increase the training speed of the model on the premise of ensuring prediction accuracy, especially when dealing with a large amount of data (such as f o F 2 data). For this reason, a modeling method using extreme learning machine (ELM) for the prediction of f o F 2 is proposed in this paper. This proposed method has fast learning speed and good generalization performance. The results show that the proposed method provides a successful prediction of the f o F 2 . Compared with the International Reference Ionosphere model, the average prediction accuracy has increased by 34%. Compared with the backward propagation neural network, the ELM model accelerates the model training speed by 61 times while maintaining the similar predicting accuracy. In summary, the modeling method using ELM for the prediction of the f o F 2 provides a new idea for the modeling problems of the ionospheric parameters. Key Points: The ELM model is applied to predict the f o F 2 for the first time Compared with the published BPNN model, the ELM model can achieve faster training process while maintaining the similar accuracy Compared with the IRI model, the ELM model has better accuracy in terms of hourly, daily, monthly, and yearly prediction of f o F 2 values