In the light of the existing control problems in coke oven gas collector pressure systems, a model predictive control (MPC) method based on subspace identification of the gas collector pressure control system is presented. Through the analysis of a number of measurable variables, the main decision variables which affect gas collector pressure including the controllable input variables and the measurable disturbance can be obtained. The proposed method employs a subspace technique for the identification of the coke oven gas collector pressure control system, and simulates the disturbance of coal charging by a pulse signal with a certain width. A two-layer structure optimal control system of coke oven gas collector pressure is established and thus steady-state optimization and dynamic control can be realized, respectively. With consideration of the influence of measurable disturbance, we calculate the steady-state target through the degrees of freedom of the system, and then constrained MPC dynamic optimization is carried out. The algorithm has been successfully applied in the JN-8-type coke oven gas collector pressure control system of an iron and steel group, and has achieved good results.