The deformation of foundation pit slope has the characteristics of non-stationarity and non-linearity. Currently, prediction models for foundation pit slope deformation usually use the data of a single monitoring point or overall monitoring points to predict, ignoring the correlation between different monitoring points. Three models, empirical mode decomposition-particle swarm optimization-back propagation neural network(EMD-PSO-BPNN) model, PSO-BPNN model, and BPNN model were built. Those models are based on single monitoring point data and related monitoring point data. Finally, a deep foundation pit slope in Chongqing was used to verify the correctness of those models. The following conclusions can be obtained. In the first place, The EMD model reduces the non-stationarity of the deformation data of the foundation pit slope, and makes the curve of each component smooth and stable, which improves the prediction accuracy. In the second place, the EMD-PSO-BPNN model has better ability of non-linear mapping, learning and self-adaptation. The prediction accuracy of the EMD-PSO-BPNN model is better than that of other models. In the last place, under the same model, the prediction accuracy of the prediction model based on correlation points is significantly higher than that of the prediction model based on single monitoring point.