Atmospheric coherence length monitor realizes continuous observation of the star for all-weather, and it is a kind of conventional instrument to measure the coherence of the atmosphere. In the actual measurement process, the atmospheric coherence length meter instrument needs through the GPS receiver to locate the local longitude, latitude and GPS time, in order to calculate the star position in real time. When the signal conditions are not ideal, such as indoor, forest and urban environment, the phenomenon like occlusion, multipath and interference are more severe. At this point, will lead to signals of GPS receiver appear error and the GPS receiver positioning accuracy will be greatly reduced, May lead to the telescope cannot tracking stars normally. In order to improve the GPS receiver positioning accuracy, the support vector machine (SVM) theory based on statistical principle is adopted to avoid the "overfitting problem" which minimizes the risk of simple experience. Application of GPS receiver receives the data information in different environment as the training sample and test sample, Data information include longitude, latitude, and satellite elevation, satellite azimuth, and satellite signal-to-noise ratio, which may affect the accuracy of GPS positioning. Using the linearly separable support vector machine (SVM) algorithm, by finding out the maximized class distance to get the optimal classification super planar to classify the data sets. And then, enter the stage of testing, according to the positioning accuracy classification results, predict the future output data, filter the positioning accuracy of the error data.