Abstract: Based on neural networks, a method to adjust the time delay of active control was explored Choosing the inputs and outputs of the neural network properly, the network can be trained efficiently. After training, the network could predict the future response of structures by using the past history of its response. Thus, it could give some time to wait for for measuring, for transmission and for actuator motion, and the time delay of active control would be reduced or erased finally. In the end, predictive control of Naming TV Tower under wind excitation is performed.