Abstract: Due to the complexity of the mining environment and the problems of heavy measurement workload and time-delay existing in the conventional leveling，it is proposed that the support vector regression algorithm calculating is used to calculate the refining quasi-geoid of height anomaly based on the theory of statistics，and GPS height was applied to the mine fast leveling survey.The global search optimal parameter of support vector regression training based on genetic algorithms solved the human blindness in selecting parameters in regression model，and improved the generalization ability of the algorithm and the regression accuracy.Finally，the mine field data was used to compute height anomaly by contrasting with polynomial fitting and radial basis function neural network.The results showed that the support vector regression based on genetic algorithm is simple in structure，and its accuracy is better than that by polynomial fitting and radial basis function neural network.It can be applied into mine GPS elevation fitting.

Key words: GPS height anomaly, Support vector regression, Genetic algorithm, Radial basis function neural network