Prediction of Ionospheric F2 Layer Heights Obtained from HFSWR during Typhoons Utilizing GA-BP Neural Network Model

The ionosphere which is an essential structure of the Earth’s atmosphere is important in wireless communication and radar positioning. The F2 layer height parameter of the ionosphere is subject to variations by severe convective weather such as typhoons. The prediction of the ionospheric F2 layer height data obtains the height variations over a period of time, and subsequently the forecasted ionospheric F2 layer height parameters can be further analyzed to derive the typhoon wind speeds and intensities, which are very valuable for reducing the loss of life and property in the coastal areas. Therefore, the accurate prediction of the ionospheric F2 layer height is of great research significance. This study adopted a BP (back propagation) neural network optimized by the genetic algorithm (GA-BP) to forecast the ionospheric F2 layer height data that were obtained by high frequency surface wave radar (HFSWR) during the typhoon. The genetic algorithm was applied to optimize the weights and thresholds of the BP neural network to obtain better predictive performance. Subsequently, four criteria, root-mean-square error (RMSE), absolute-mean error (MAE), absolute-median error (MAD) and R-Squared coefficient were utilized to evaluate the proposed prediction model. The proposed GA-BP prediction model was compared with the commonly utilized BP, LSTM (long short-term memory), and CNN (convolutional neural network) models in this research. The RMSE values of GA-BP, CNN, LSTM and BP models were 0.6059, 2.1860, 2.3372 and 1.2192 respectively. The error results indicated that the GA-BP neural network proposed in the study provided the best predictive performance, and the prediction error of the model was reduced by 50% compared to the error of the BP neural network. Furthermore, among the results of MAE and MAD calculated for all the prediction models obtained, the error of the GA-BP model proposed in this study was remarkably lower than that of the other models. These error results illustrated that the GA-BP prediction model proposed in this paper was capable of accurately predicting the ionospheric F2 layer height. This paper achieved the precise prediction of the ionospheric F2 layer height with GA-BP neural network on the basis of HFSWR ionospheric F2 layer height data.