Traditional gray G1Vl(1, 1) prediction model for non一negative incxemental time series have good prediction accuracy, but the prediction accuracy is not enough for large amplitude fluctuation time series. In order to improve prediction accuracy, the trend curve- self- memory combination prediction model is es- tablished based on the trerxl curve model and the self- memory theory, and the model is applied in the prediction of the amount of displacement of deep foundation pit,and with grayCM(1, 1) prediction model and trend curve prediction model comparison, the results show that the model has high prediction accura- cy.