University of Khartoum

Geoid Determination Using Artificial Neural Networks and Geometrical Models

Geoid Determination Using Artificial Neural Networks and Geometrical Models

Show full item record

Title: Geoid Determination Using Artificial Neural Networks and Geometrical Models
Author: Aborida, Ammar Mohammed Maryod
Abstract: The determination of Orthometric heights is important for many task in Engineering, this can be determined by computing the difference between geodetic height and geoid undulation. This relationship is different and unstable from one point to another. In such trend, Artificial Intelligence proved to be useful towards finding solution for problems that could not be solved by traditional techniques. The main objective of this study is to determine an accurate geoid model for Khartoum State. The study adopts a method consisting of three sub methods to model the geoid, the first one is to apply an Artificial Neural network (ANN) to model the Geoid surface using the back propagation algorithm for Khartoum State, through supervised training by using 40 collocated points for training and 6 for test, the second method used kriging to model the Geoid surface for Khartoum State, and finally Compute the coefficients representing the bias∝_0, and tilts of the geoid plane will respect to WGS84 ellipsoid ∝_1 ∝_2 to model the Geoid surface for Khartoum state, and then comparison between these three methods with Earth Gravitational Model 2008. Based on the test results and the statistical analysis carried out in this study a trained Artificial Neural Networks model was found and is tested to be able to estimate Geoid model better than the models generated by interpolation technique (kriging method); and better than coefficients representing the bias and tilts of the geoid plane, and EGM 2008 when applied on Khartoum locality. The average of the discrepancies of the test points calculated were -1.1 cm, -0.7 cm, -0.5 cm, and 28 cm, respectively, for the modeling by ANNs, interpolation, geometrical, and by EGM 2008. The standard deviation of the discrepancies of the test points was ±1.3 cm, ±3.3 cm, ±3.7 cm, and ±4.6 cm, respectively, for the modeling by ANNs, interpolation, geometrical, and by EGM 2008. Thus, the Root Mean Square Errors of the discrepancies of the test points was computed to be ±1.6 cm, ±3 cm, ±3.7 cm, and ±28 cm, respectively, for the modeling by ANNs, interpolation, geometrical, and by EGM 2008, where means that the ANN model is better than the other three models. The study recommends running further studies using other algorithms to train the Artificial Neural Networks.
URI: http://khartoumspace.uofk.edu/123456789/27020


Files in this item

Files Size Format View

This item appears in the following Collection(s)

Show full item record

Share

Search DSpace


Browse

My Account