Predicting TEC in China based on the neural networks optimized by genetic algorithm

Abstract

This paper illustrates the application of neural networks (NN) in developing a regional prediction model for the ionospheric total electron content (TEC) over China. To avoid the ‘local minimum’ effect caused by the traditional NN-based model, genetic algorithm (GA) is utilized to optimize the initial weights of NN. In this study, the NN has 19 input parameters which are known to cause variations in the ionospheric parameters. These parameters relate to the ionospheric diurnal variations, seasonal information, solar cycle, geomagnetic activities, geographic coordinates, and declination. The output parameter is the daily hourly vertical TEC (VTEC) measured from 43 permanent GPS (Global Positioning System) stations in China. Datasets for 2012–2014 are used to train the network, and datasets for 2015 are selected as the test data to verify the model’s performance. Predictions from the GA-based NN (GA-NN) model, back propagation-based NN (BP-NN) model, and International Reference Ionosphere 2012 (IRI2012) model are then compared with the observed TEC from 12 GPS stations in China. According to the numerical analysis, the root-mean-square error (RMSE) of GA-NN model ranges from 5.2140 to 8.4756 TECU, the corresponding percent deviation (PD) is 8.78–12.30%, and the correlation coefficient falls within the range of 0.8069–0.9583. The BP-NN model’s RMSE varies between 6.2962 and 12.1468 TECU, PD is between 10.17% and 14.16%, and the correlation coefficient lies in the range of 0.7192–0.9348. For the IRI2012 model, the corresponding ranges are 6.5513–11.7937 TECU, 11.01–14.07%, and 0.7292–0.9129, respectively. This, combined with the comparison of diurnal variations of TEC, suggests that the GA-NN model greatly outperforms the BP-NN and IRI2012 models. Furthermore, the variation of seasonal and local characteristics are also validated by the GA-NN model. The results indicate that the GA-NN model is very promising for applications in ionospheric studies.