Solvent-based post-combustion CO2 capture plant has to be operated in a flexible manner because of its high energy consumption and the frequent load variation of upstream power plants. Such a flexible operation brings out two objectives for the control system: i) the system should be able to change the CO2 capture rate quickly and smoothly in a wide operating range; ii) the system should effectively remove the disturbances from power plant flue gas. To achieve these goals, this paper proposed a multi-model predictive control (MMPC) strategy for solvent-based post-combustion CO2 capture plant. Firstly, local models of the CO2 capture plant at different operating points are identified through subspace identification method. Nonlinearity analysis of the plant is then performed and according to the results, suitable local models are selected, on which the multi-model predictive controller is designed. To enhance the flue gas disturbance rejection property of the CO2 capture plant and attain a better adaption to the power plant load variation, the flue gas flow rate is considered in the local model identification as an additional measured disturbance, thus the predictive controller can calculate the optimal control input even in the case of flue gas flow rate variation. Simulation results on an MEA-based CO2 capture plant developed on gCCS show the effectiveness and advantages of the proposed MMPC controller over wide range capture rate variation and power plant flue gas variation.