Abstract: In order to improve the accuracy and reliability of GPS height anomaly fitting,the GPS height anomaly fitting method based on the relevant vector machine (RVM) is established.The new model has the characteristics of sparse and probability.The relevance vector machine is established based on cauchy kernel function and the cross validation method,and the formula of confidence interval estimation is established.Taking the GPS height control network in a mining area as an example,the GPS height anomaly fitting model based on the relevant vector machine is put forward，the height anomaly fitting accuracy of the polynomial,BP neural network,genetic algorithm least squares support vector machine and the relevant vector machine are analyzed in depth.The reliability of the above methods are evaluated by means of confidence interval estimation.Indicators of mean absolute error(MAE),mean absolute percentage error(MAPE) and root mean square error(RMSE)are adopted to conducted evaluation of the accuracy of the above methods.The research results show that:①values of MAE,MAPE,RMSE of the new method proposed in this paper is superior than polynomial,BP neural network and genetic algorithm least squares support vector machine;②GPS height anomaly values of test data set are all within the estimated confidence intervals.The above research results further indicated that the fitting precision and reliability of the GPS height anomaly based on the relevant vector machine are good,it is very suitable for the GPS height anomaly fitting,therefore，the normal height in mining area can be measured quickly with high precision by this model.

Key words: Height fitting of mining area, Height anomaly, Polynomial fitting, BP neural network, Genetic algorithm least squares support vector machine, Relevant vector machine