A model prediction controller (MPC) based on radial basis function (RBF) neural network is designed to counter the model uncertainty and multiple constraints of the autonomous underwater vehicle (AUV). On this basis of path following control with MPC, the RBF neural network is trained online with real-time measurement data to compensate for the AUV's model uncertainty, thus suppressing the interference of model uncertainty on the MPC and reducing its overshoot and tracking error. Simulation results show that the path following algorithm based on RBF-MPC has better transient and steady-state performance compared with the classical MPC algorithm. 针对自主水下航行器(AUV)的模型不确定性和多约束的特点, 设计了基于径向基(RBF)神经网络的模型预测控制器。在使用模型预测控制(MPC)进行路径跟踪控制的基础上, 利用实时测量数据在线训练RBF神经网络, 对AUV模型不确定性进行补偿, 抑制了模型不确定性对模型预测控制器的干扰, 减小了系统的超调量和跟踪误差。仿真结果表明, 基于RBF-MPC路径跟踪控制算法与经典的MPC算法相比, 具有更好的暂态和稳态性能。