Abstract: On-line identification and self-tuning prediction of the ARMA under the influence of stochastic distribution sequence with time varying mean and time varying variance have been studied. The on-line identification of the model can be achieved by a simple ELS method. The self-tuning prediction of the future value can be achieved on the basis of real time model-building and modification to the one-step predicting errors sequence in the exponential smooth predictor. It can be used for short period prediction of low altitude flight data by means of Rader measuring. Simulation and processed results of the measuring data show that the method is simple and can acquire high precision. It can be used for real time model building and short period prediction to some kind of non-stationary process via microcomputer.