A novel distributed model predictive control (DMPC) approach based on immune algorithm (IA) to find out the optimal system decomposition structure is proposed. The IA is used to solve decomposition problems for input clustering decomposition (ICD) and input?output pairing decomposition (IOPD), which can minimize the impact of input?output coupling between systems, and then DMPC algorithm is used to control the decomposed system. This approach effectively reduces the coupling between subsystems, and reduces the communication load of the system. Finally, a heavy oil fractionation chemical process is simulated and compared with the centralized MPC simulation results to verify the effectiveness of the algorithm.