用神经网络进行结构响应和控制信号的时滞补偿 ,讨论了基于多层前馈网络的预测控制算法对多层结构进行振动响应控制。为使神经控制器在有时滞的情况下有更好的补偿效果 ,提出了在控制器的训练中引入神经辨识器的训练方法。并且在控制准则中考虑动作器的容限 ,在控制算法中将控制目标定义为逐步减小未来几个时刻响应的平均值而不是每一个时刻的响应 ,使结构的受控响应更加平滑 ,改善了由于时滞效应引起的控制失稳问题。仿真结果表明 ,该算法有较好的控制效果。 The neural network is used to carry out the structural response and control signal delay compensation. The prediction control algorithm based on multi-layer feedforward network is discussed to control the vibration response of multilayer structure. In order to make the neural controller have a better compensation effect under the condition of time lag, a training method of introducing the neural identifier into the controller training is proposed. In the control criterion, the tolerance of the actuator is considered. In the control algorithm, the control objective is defined as gradually reducing the average value of the response in next few moments rather than the response of each moment, so as to make the controlled response of the structure more smooth and improve Due to the lag effect caused by the control of instability. Simulation results show that the algorithm has better control effect.