Research on regional ionospheric TEC modeling using RBF neural network

Abstract

Machine learning techniques which are about the construction and study of system that can learn from data are combined with many application fields. A method on ionospheric total electron content (TEC) mapping is proposed based on radical basis function (RBF) neural network improved by Gaussian mixture model (GMM). Due to the complicated ionospheric behavior over China, GMM is used to determine the center of basis function in the unsupervised training process. Gradient descent is performed to update the weights function on a sum of squared output error function in the supervised learning process. The TEC values from the center for orbit determination in Europe (CODE) global ionospheric maps covering the period from 2007 to 2010 are used to investigate the performance of the developed network model. For independent validation, the simulated TEC values at different latitudes (20°N, 30°N and 40°N) along 120°E longitude are analyzed and evaluated. The results show that the simulated TEC from the RBF network based model has good agreement with the observed CODE TEC with acceptable errors. The theoretical research indicates that RBF can offer a powerful and reliable alternative to the design of ionospheric TEC forecast technologies and thus make a significant contribution to the ionospheric modeling efforts in China.