Abstract: The main problem for the application of nonlinear model predictive control is to solve the nonconvex in finite sampling time.Traditional method has some defects such as high computation and easy to fall into local solution.It is a wise choice to solve this problem by using intelligent method.In this paper,some modifications of the traditional differential evolution(DE) algorithm are made,including accommodating main parameters to increase the convergence rate,and employing various mutation strategies to increase multiplicity of the population and avoid falling into local solution.The simulation results in the three-tank system show that it is effective and available,with a good feature for application in industry.