In order to get accurate results of the foundation deformation prediction under the conditions of "limited sample" and "limited information", the research of Gray BP neural network combined model is proposed, based on the traditional models including GM (1,1) model and BP neural network model. By summarizing the two traditional models' principles and algorithms, concluding their respective advantages and disadvantages and analyzing the nature of the relationship between the models, the way of structuring the combined model is given out. Finally, the test of Gray GM (1, 1) model, BP neural network model, Gray BP neural network combined model, by using of the monitoring data of theYantang station on No.3 Line of GuangzhouMetro, confirms the superiority of the combined model.