Abstract: In order to improve the accuracy of deep foundation pit deformation prediction, a feedback Elman neural network model was introduced into the foundation pit surface settlement prediction. The rolling prediction of foundation pit settlement displacement time series can be realized by using Elman neural network algorithm. Taking the foundation pit engineering of a station of Xi'an Metro Line 5 as the example, based on the idea of combined forecasting, combining the two forecasting methods of Elman neural network and Markov chain, the prediction model of ground subsidence of foundation pit was established. The random disturbance error was corrected by Markov chain model. The effect of Elman neural network prediction model before and after the modification were both better than the BP neural network prediction model. Graphical User Interface (GUI) prediction system based on MATLAB was designed and developed, making the model prediction process easy and convenient and the prediction process can be displayed dynamically by graphical results.