中文摘要：

当基坑水平位移数据呈现某种趋势时，常常需要将这种有趋势的时间序列分解为趋势项与随机项，建立指数平滑为趋势项的自回归模型，进行分析、预测。工程实例验证表明：该方法与单独指数平滑模型相比，预测值与实际位移值之间的误差更小；与单独的GM（1，1）模型相比，拟合线形更加稳定。

英文摘要：

When the excavation foundation ditch horizontally deforms, its dyanmical process of the horizontal displacement is frequently decomposed into two parts with stochastic noise. The deformation is analyzed and predicted by autoregressive model whose trend is exponential smoothing model. The engineering experiment shows that the method has less predication error than the single exponential smoothing model, and the fitted curve is more stable than that of GM （ 1,1 ） model.