Predictive control algorithm based on the step or impulse response of the object, using a rolling over method realizing process on-line optimization control. In the optimization process, it gets the feedback correction continuously through the comparison between the measured system output and prediction model output, it is able to overcome certain extent impact as a result of prediction model error structure disturbance and some state of uncertainty noise disturbance. There are lots of methods of predictive control, the predictive control can deal directly with a pure lagged object and has a strong adaptability of large inertia and a good tracking performance and robustness, this paper analyzed the stability and robustness of the dynamic matrix control, this response can directly access from the industrial scene, it does not require a priori knowledge of the model structure, so we can design control systems without going through the complex process of identification.