Abstract：The principle of self-adaptive time series model is introducing filter theory to time series analysis. The self-adaptive time series model can，to some extent，modulate model parameters according to the input and output data，and adjust them to its optimal values in some statistical conditions through the iterative algorithm. The satisfying forecasting results may be obtained even if only part of the statistical properties is known，which is proved by two examples of modeling and forecasting of the convergent displacements of tunnels. The presented model is also useful to forecast the stability of surrounding rock and support structure of tunnels.
