Abstract:

The problem of transformation between geodetic ellipsoidal coordinates (φ, λ, h) and three-dimensional (3D) global Cartesian coordinates (X, Y, Z) is a common one. Only one method for solving this transformation problem has been found. An examination of the current studies reveals that the application and testing of possible alternative techniques for transforming geodetic ellipsoidal coordinates to 3D global Cartesian coordinates have not been fully addressed. The aim of this study was to investigate the performance of a method of transforming geodetic ellipsoidal coordinates to 3D global Cartesian coordinates using a multilayer perceptron artificial neural network (MLP-ANN). For the prediction, 594-point datasets of the Turkish National Fundamental GPS Network (TNPGN) were used. As a result of many trials, the Bayesian Regulation was determined as the training algorithm and the number of hidden layers was 2. The mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R2) were used for the performance evaluation of the MLP-ANN model. According to the test results, the MSE value for the components of the MLP-ANN and the 3D global Cartesian coordinates varied between 0.4536 cm and 0.9411 cm, the MAE value ranged between 0.3883 cm and 0.8165 cm, and the R2 value was calculated as 0.9999 for all models. To examine the results in more detail, the difference between the estimated values and the calculated values was determined. Accordingly, it was found that the difference in the values was quite close to zero except for a few number of data. According to these statistical criteria, it was concluded that the MLP-ANN used in this study could be used as an alternative to the classical coordinate transformation method.

Keywords: