This paper presents a least-square support vector machine based linear parameter-varying model approach for multiple-input multiple-output nonlinear system (MIMO-LSSVM-LPV). The identified model can be used in the model predictive control scheme combined with genetic algorithm (GA-MPC). The new identification and controlling integration scheme is named MIMO-LSSVM-LPV+GA-MPC. Simulation results show that the identification algorithm can approximate complex nonlinearity with high accuracy while keep the advantage of low computational burden of linear regression. GA based MPC can get the real-time optimal control input and achieve good controlling performance.