Investigating the Efficiency of Deep Learning Methods in Estimating GPS Geodetic Velocity

Abstract

Geodetic velocity (GV) has many applications in tectonic motion determination and geodynamic studies. Due to the high cost of global navigation satellite system stations, deep learning methods have been investigated to estimate GV. In this research, four methods of convolutional neural networks (CNNs), deep Boltzmann machines, deep belief net and recurrent neural networks have been applied. The GV of 42 global positioning system stations is entered the deep learning methods. The outputs of the four methods have successfully passed the normality test. The results show that the CNN method has a lower goodness of fit and root mean square error (RMSE). CNN can learn different dependencies and extract features.