A full data-driven method of closed-loop subspace identification and predictive controller design is presented, in which the subspace matrix is completely identified through the input and output data of the closed-loop system. By splitting the subspace matrix and eliminating the input of the model correlated to disturbance, the unbiased estimators of the parameters of the subspace matrix are obtained. The predictive controller is designed by using the closed-loop system subspace matrix which is obtained by identifying directly as the forecasting model. The control method is applied to the pressure regulating system for the coking chamber of an iron and steel group, and good performence is obtained.