Abstract: To calculate control sequence,nonlinear model predictive control(NMPC) can be tranformed into an optimization process of nonlinear programming problem with constraints.In this article,three strategies to consider constraints are analyzed.According to the features of genetic algorithm(GA),state vector are calculated by state(equation,) in which the dimension of search space is reduced as well as the equation constraints which are difficult to deal with by genetic algorithm are eliminated.The results of both simulation and practice to a coupled-tank apparatus illustrate that NMPC based on GA outperforms which based on sequential quadratic programming(SQP).To avoid much time consumptions of GA and randomicity of optimization results,a hybrid method is proposed by incorporating SQP.The results of both simulation and practice verify the feasibility and efficiency of proposed method.