Abstract: A multiple flexible beams coupling structure is investigated in this study, and an experimental platform of multiple flexible beams coupling structure is built. The laser displacement sensor is used to measure the residual vibration signal, and the finite element model is established. A neural network controller is designed to suppress the residual vibration, including prediction and control networks. The prediction network is long short-term memory (LSTM) neural network, and the control network is a backpropagation neural network. After off-line training, the network uses backpropagation to update the weights. The simulation and experimental results show that the LSTM network has a better predictive performance, which confirms the effectiveness of the neural network controller. In addition, it can suppress vibration faster.