Abstract: An integrator is usually utilized in a process controller so as to attain the zero steady state error. In order to consider the integral effect, this paper utilizes incremental inputs to replace control inputs in LQG controller’s design for the objective function of the infinite horizon MPC. A novel LQG benchmark computing method based on the subspace is designed for MPC performance analysis. Both subspace matrices and noise variance are estimated by using closed loop data and subspace identification algorithm. And then, the optimal LQG benchmark variances can be computed. Simulations via Wood Berry binary distillation column demonstrate the effectiveness of the proposed method.