Abstract: In order to find out the intrinsic regularities from enormous observed data in TGP's safety monitoring, a mathematic model which can forecast the stress strain status at key construction positions would be established. An "explorative data" analytic method (analog optimization method) was used for studying the structure of forecasting model.To take the displacement of the first seepage prevention wall of TGP's upstream cofferdam as illustration,three kinds of forecasting models were built respectively by multiple regression analysis,grey model and time series analysis. Comparison between calculated results by these models and easured values showed that the residual error of 1st model is rather small, that of the 3rd model next;for the fitting degree of models, the 1st model is rather better, 3rd model next and the 2rd model worse.The final detection of measured data showed that the effect of 1st model is still better, the reliability of forecast representing measured data in the forecasted range is 95%. So, the multiple nonlinear regression model is suggested to be selected as the forecasing model.