For the characteristics that multi-variable predictive control has a large amount of calculation, and its control effect is sensitive to disturbance and model mismatch, a feedforward decoupling strategy based on a control model for the engineering application of multi-variable predictive control is proposed. Based on structure analysis, the important pairings between controlled variables and manipulated variables are preserved, and others are introduced as feedforward to make up the influence, which simplifies the system structure, reduces the coupling between controlled variables and manipulated variables, weakens the sensitivity of the predictive controller to the disturbance and model mismatch. The single input single output or smaller multiple input multiple output system formed under extreme circumstances reduces the amount of online computing. Then a decentralized optimization strategy based on distributed predictive control is proposed to reduce the system scale and online calculation amount further. Finally, the simulation examples are given to verify the feasibility and effectiveness of the proposed method.