提出采用多层局部回归神经网络建立多变量非线性系统多步预测模型的方法，神经 网络模型可提供多步预测控制所需要的系统输出预测值及输出向量对控制向量的雅可比矩 阵．仿真试验表明这种动态神经网络的预测模型具有较高的精度． A multi-layer local regression neural network is proposed to establish a multi-variable nonlinear system multi-step prediction model. The neural network model can provide predictive value of system output and Jacobian matrix of output vector to control vector required by multi-step predictive control. Simulation experiments show that the prediction model of this dynamic neural network has higher accuracy.