提出一种基于神经网络的非线性多步预测控制。采用由线性网络和动态递归神经网络构成的复合神经网络，在此基础上将线性系统的广义预测控制器扩展为非线性系统的多步预测控制器。通过对非线性过程ＣＳＴＲ的仿真表明，该方法的稳定性和鲁棒性明显优于线性ＤＭＣ预测控制。 A nonlinear multi-step predictive control based on neural network is proposed. A composite neural network consisting of a linear network and a dynamic recurrent neural network is adopted, on the basis of which a generalized predictive controller of a linear system is extended to a multi-step predictive controller of a nonlinear system. The simulation of nonlinear CSTR shows that the stability and robustness of this method are obviously better than that of linear DMC predictive control.