Research On The Application Of GPS Elevation Conversion In Road Vertical Control Survey

The road vertical control survey plays a very important role, providing a uniformelevation benchmarks, in the design construction and operation stage. With the developmentof GPS (Global Positioning System), GPS technology be used to in the road vertical controlsurvey becomes inevitable, that can reduce work load and bring direct economic benefits.GPS measurement with the a lot of advantages such as high efficiency, high precision, free ofthe climate condition limit, The geodetic height of observation point be gained by relativepositioning technology of GPS, But the normal height is most commonly employed as high ofthe ground point. GPS elevation conversion was researched based on neural network and leastsquare collocation method with the data of road vertical control survey. The mainachievements was made in the following:(1)The relationship between the height system and the GPS technology principle wereunderstood on a large number of data. Research status of GPS elevation conversion be knownin road vertical control survey. The accuracy requirement and measuring method wereunderstood in road elevation control survey.(2) The normalization of data is inevitable, preventing the large difference numericaltransformation parameters, in order to making its have equal weight based on the model ofneural network GPS elevation Conversion. The traditional normalization function (function A)which domain is [0,1], In order to research the data normalization way for GPS elevationconversion of the outcome, this paper puts forward normalization function (function B) whichdomain is [1,1].The Influence of normalization function was discussed and Combined withthree neural network model, such as back propagation (BP), generalized regression (GRNN)and radiate basis function (RBF) and so on. The results show that the normalization functionB is better than the normalization function A. when the normalization function B is adopted,the result of neural network (GRNN) the GPS elevation conversion is best.(3)The least squares collocation model with the random vector parameter was putforward as the random vector parameter did not take into account in Conventional modelabout GPS elevation transformation. Compared the least squares collocation model with theplane conversion model considering the measured data, relative accuracy increasesignificantly. The result of the error is0.039m when the plane the transformation model isused. However, the result of the error is0.008m when the least square collocationtransformation model is used. Gauss function, Hirvonen function and Gaussian function ascovariance function be calculated and analyzed in the GPS height of transformation Respectively. When the gaussian function is deemed to covariance function, that result isbetter than before both.(4)Influence factors of precision which GPS elevation be converted to normal high inroad vertical control survey were discussed. There are a lot of errors in the process ofobservation, such as the error caused by satellite star calendar, the error of GPS signaltransmission, the error of GPS antenna high measurement, the error of bench mark, the errorcaused by the league and so on. There are many errors about the model of GPS heighttransformation, such as model range, the choice of model parameters, the structure of GPSnetwork, the choice of known points and so on.The feasibility and reliability of GPS elevation conversion in the road vertical controlsurvey be verified by means of the example of GPS elevation conversion was calculated bythe generalized regression neural network (GRNN).