Multi - parametric evaluation of Back Propagation Artificial Neural Network in Geoid Undulations determination modeling
20/09/2016 | 16:45 | Session 3: Local/regional geoid determination methods and models
Author(s): George Pantazis
The increasingly use of GNSS for technical constructions and infrastructure works makes necessary the calculation of accurate geoid undulation’s models or maps for determining accurate orthometric heights. There are different methods for geoid height determination. Artificial neural networks (ANN) are also used for this purpose. Specially, back propagation artificial neural networks (BPANN) are widely applied for engineering practice. It is well known that the training of an ANN consists a “black box”, as the user cannot interlope in the procedure. The aim of this work is to investigate some parameters and the grade of their influence to the results as well as to their accuracy, when the calculation of geoid undulations is carried out by a BPANN. Parameters as different types of input data (ellipsoidal or Cartesian coordinates, components of the deflection of the vertical), the density of the known points at a concrete area’s size and the points’ weight variation are examined. The results, RMSEs and the shape of outcome surfaces of geoid undulation are compared to each other and to the initial calculated surface by a polynomial interpolation method using GPS/leveling data. For the trial run 37 points of known geoid undulations, dispread at 12Km2 area, of an urban region of Athens city were used. Useful conclusions are illustrated in diagrams in order to evaluate the BPANN’s use for the geoid undulation determination.