SUMMARY
In this study, test-region global positioning system (GPS) control points exhibitingknown first-order orthometric heights were employed to obtain the points of planecoordinates and ellipsoidal heights by using the real-time GPS kinematicmeasurement method. Plane-fitting, second-order curve-surface fitting, back-propagation (BP) neural networks, and least-squares support vector machine (LS-SVM) calculation methods were employed. The study includes a discussion on dataintegrity and localization, changing reference-point quantities and distributions toobtain an optimal solution. Furthermore, the LS-SVM was combined with localgeoidal-undulation models that were established by researching and analyzing3kernel functions. The results indicated that the overall precision of the localgeometric geoidal-undulation values calculated using the radial basis function(RBF) and third-order polynomial kernel function was optimal and the root meansquare error (RMSE) was approximately ± 1.5 cm. These findings demonstrated thatthe LS-SVM provides a rapid and practical method for determining orthometricheights and should serve as a valuable academic reference regarding local geoidmodels.