Research On The Method Of GPS Elevation Abnormal Fitting Based On Artificial Neural Network

The base line vector between two points and precision geodesy height difference can be acquired after the processing of GPS observing data, if one of the points of geodetic height is known, then all the points geodetic height in the net can be acquired. Geodetic height is the system which referenced to ellipsoidal surface, however, in our country the height system adopted in engineering is normal height which referenced to quasi-geoid. Because the quasi-geoid is an irregular curved surface, it is unable with an essence accurate curved surface to simulate, this enables GPS only to provide for us the accuracy of the geodetic height, but is not the normal height in our engineering needs. So it has seriously affected the three dimensional localization application develop of GPS and makes the superiority of providing the three dimensional coordinates not be able to obtain the full display.So, the precision height resources of GPS usually be wasted in production and isn't make good use and not be fully developed. If we give up the use of precision geodesy height difference of GPS surveying, then the superiority of GPS measure was not fully bring into play. Because GPS leveling is much easier than traditional leveling, so people expect GPS leveling would gradually take place of traditional leveling. But, the engineering hoped to unify all the result to station normal height. Then, we should study the transform between GPS geodetic height and normal height, thus came into the problem that the elevation abnormal should be studied.At present, many domestic and foreign scholars are studying the problem of GPS elevation abnormal fitting, from the view point of comprehensive all the methods of GPS elevation abnormal fitting and it can be divided into three types: geometry analytical method (mathematical model method),physical geodesy method,neural network method. Based on the analyses in the last few years research results of the domestic and foreign GPS leveling, this paper studied the methods on GPS elevation abnormal fitting based on Artificial Neural Network; it has done some jobs and the results as follows:(1)It analyzes and summaries the latest research results of GPS elevation abnormal fitting,and based on this, it discusses the main features of BP neural network of GPS elevation conversion method and analyses BP algorithm mathematical models and network architecture;structures the GPS elevation abnormal fitting model based on BP neural network,second surface fitting and surveying area gravity;finally it realizes the GPS elevation fitting procesures based on neural network.(2)It combinates the gravity method and BP neural network to realize the GPS elevation conversion,the result shows that the method has advantages of net stable, conversion accurate to the single BP neural network,and so on.(3)It introduces the Normalization of Bayesian algorithm based on M×N network to GPS elevation conversion,and the results has high quality,stable net and more real time.(4)It proposes the combination method of second surface fitting algorithm and Radial Basis Function Neural Network to realize the GPS elevation conversion, and the method has high fitting accuracy, strong web and fast calculation. Through the experiment, the net is the most optimal.(5)Introduces the theory of Akaike Information Criteria, and then applies it to the optimization of the GPS elevation abnormal transform model, which supplies a theory to Radial Basis Function Neural Network model's optimization and overcame the weakness of certain artificial neural network training parameters.(6)It studies the BP network model, Radial Basis Function Neural Network model, BP neural network combination algorithm and Radial Basis Function Neural Network based on engineering data, through these comparision and analysises to the fitting models it proposes the accuracy assessment to the diverse fitting models.