This study intends to integrate the existing 1st order leveling data, GPS coordinates and ellipsoidal height to find out the geoidal undulation model of Taichung city. The main simulation is done by adopting Particle Swarm Optimization to better fit local geoid by traditional quadratic surface. In accordance with the experiment, we propose an improved result by using the Particle Swarm Optimization and showing a computed Root Mean Square Error about ± 1.02cm without imposing any parametrical restrictions.