Research On GPS Height Fitting Based On SVM

Currently, the surveying technology by using GPS(Global Positioning System) is becoming one of the most widely used surveying measure in our daily surveying work, the technology has a high-precision of three-dimensional coordinate positioning, automatic measurement, all-weather and real-time positioning and other prominent features. The reference plane of the surveying technology by using GPS is the surface of WGS-84 reference ellipsoid. The obtained elevation by using GPS technology is geodetic height. The elevation used in the engineering survey is normal height in our country. If we can accurately and quickly transforms geodetic height into normal height, it can replace a considerable part of traditional leveling work, and will have a very large economic benefit. Currently, the direct way to obtain normal height by using GPS is that we can convert geodetic height obtained by GPS to normal height by using the precise height anomaly which is obtained by refined quasi-geoid. However, to obtain a highly accurate quasi-geoid require high precision GPS data, gravity observation data, DTM data and earth gravity model, it is a big complicated project. A small company or a small region does not have the above conditions, so GPS height fitting is still a convenient, fast, economical way to obtain normal height by using GPS.This paper introduces the basic concepts of height system and principles of Leveling, and describes the purpose and significance of GPS Height Transformation, then the current mainstream GPS Height Transformation methods are discussed. We can obtain normal height in direct method or fitting method. I study the principle of these methods in-depth, and give a summary of domestic and abroad common methods, then analysis and discuss the scope of these methods. The basic principle of SVM(Support Vector Machine) and its application in GPS height fitting has been introduced in detail in the paper. In solving small sample, nonlinear problems,SVM has many unique advantages. SVM applied to GPS height fitting can make up for defects of the lack of data. Using MATLAB software programming, SVM theory has been applied to the GPS height fitting instance. By the comparison of BP neural network, I find at different distribution of GPS elevations/leveling points that SVM can obtain a better GPS height fitting results. Several parameters optimal selection method has been introduced for the support vector regression model. By a GPS elevation height test, these methods were compared and analyzed. Specific conclusions and suggestions has been given in the end, and it has a guiding significance in leveling.