Abstract:Aiming at the big model error of least squares support vector machine(LSSVM) in regional quasi-geoid fitting, we establish the combination model of LSSVM and Shepard interpolation model for GPS elevation conversion. We use LSSVM to fit the medium-long wave term in abnormal height, and use Shepard interpolation model to generalize and remove the residual of medium-long wave term. Combined with engineering examples in plain area and plateau mountainous area, we use the quadric surface model, the LSSVM, the Shepard interpolation model, the quadratic-Shepard model, and the LSSVM-Shepard model for elevation conversion and accuracy comparison. The results show that the accuracy of new combined model is higher than that of single model. Moreover, the conversion effect of LSSVM-Shepard model in the plain area is basically consistent with the quadratic-Shepard model, and the fitting effect is better than the quadratic-Shepard model in the plateau mountainous area.