Abstract:This paper proposes the BP neural network method which is based on particle swarm optimization algorithm to refine the regional quasigeoid, it by selecting the neural network weights and thresholds reasonably can effectively avoid the defect that the network slow convergence and easily trapped into the local optimal. This method is verified through the data of Poyang Lake and the results show that PSOBP neural network model can still obtain a good effect in the compared with the two surface partition fitting method absence of considering terrain conditions, which is selected according to the height anomaly variation.