Ionospheric Prediction Based On Deep Learning Neural Network

The ionospheric F2 critical frequency(foF2)and total electron content(TEC)are two important characteristic parameters of the ionosphere.The prediction of foF2 and TEC is of great significance to radio wave propagation,satellite navigation and short-wave communication systems.This paper mainly consists of two parts.The first part is ionospheric foF2 prediction based on traditional neural network algorithms,and the second part is ionospheric TEC short-term prediction based on deep learning neural network algorithm.In the first part,foF2 data of a solar activity cycle from 1990 to 2003 are selected from nine vertical stations of the China Radio Propagation Research Institute.In order to improve the accuracy of ionospheric foF2 prediction based on Back Propagation(BP)neural network,we adopt an improved particle swarm optimization neural network method to optimize the initial weights of BP network to prevent local optimum in the training of neural network in.By comparing the prediction results of neural network based on particle swarm optimization with those of neural network optimized by genetic algorithm,we find that the two methods have good performance for BP neural network.Compared with the results of the International Reference Ionosphere Model(IRI2016),the results show that the proposed adaptive mutation particle swarm optimization neural network can effectively improve the prediction accuracy of foF2,and has better prediction effect in low latitude area.The second part chooses the TEC data provided by International GPS Service(IGS)and IRI 2016 model.We constructed a deep learning neural network model for predicting 2 hours,6 hours,12 hours,18 hours and 24 hours ahead of calm time and storm time for the serialized TEC maps in China and Europe.By comparing the prediction results of the model with the TEC data provided by the IRI model,we find that the effect of using the deep learning neural network model in short-term prediction is better than that of IRI 2016.Combining the results of the two regions,we conclude that the model based on the deep learning neural network algorithm can effectively improve the prediction accuracy of TEC map,and has a good prediction effect in the period of weak magnetic storm.