Abstract: According to the existing problems of the conventional method of dam displacement prediction， BP neural network based on improved particle swarm optimization (IPSO BP) is put forward to predict the dam displacement. The weights and threshold of the conventional BP neural network are optimized by IPSO, thus making up the shortage of BP network and improving the prediction accuracy. The observed longitudinal displacement of the typical section of a concrete gravity dam crest from January 10， 2012 to July 31， 2012 is taken as the research object. And based on the IPSO BP prediction model the simulation analysis is carried out. At the same time， in order to verify the effect of fitting and prediction of the model， a statistical model using the least squares method to make parameter analysis and PSO BP model are developed. It is found from the prediction accuracy given by this model that the model is superior to the conventional model； the fitting effect is the best; and the average relative error of the prediction results is the minimum. Therefore, it shows that this method is effective and feasible in prediction and analysis of dam engineering.