With respect to the auto-coorelation existed in sampled data, a dynamic process modeling method based on auto-regression(AR) model is proposed. Firstly, orthogonal signal correction(OSC) is used to eliminate the orthogonal uncorrelated components from the two datasets used for regresing AR model. Then, the partial least square(PLS) is employed to get the AR model on the basis of preprocessed process datasets. The latent components extracted by the regression model as well as the residuals are together monitored online. With the involvement of Bayesian inference, the resulted multiple monitoring statistics are combined into a single probabilistic index, and triggering fault alarm can thus be simplified. The simulation experiment on the Tennessee Eastman(TE) benchmark process demonstrates the effectiveness of the proposed method.