Based on the monitored extreme stress of bridges,a combinational forecasting model of extreme stress isconstructed with the Bayesian dynamic linear model ( BDLM) .Then,the proposed model is used to forecast thestructural reliability indices in real time.Moreover,a project case is provided to analyze the forecasting precisionsof the combinational forecasting model and the existing single forecasting models.The results show that,though theproposed model is of an one-step forecasting value close to that of the single forecasting models and the mean pointforecasting model,it is of higher forecasting precision,and that,as compared with the deterministic reliability indicesfor extreme stress monitoring,the combinational forecasting model fully considers the uncertainty and randomnessof monitored stress and helps to obtain smaller reliability indices,so that it is more effective in forecasting thestructural safety.