Piezoelectric sensors or actuators are bonded to the surfaces of a flexible cantilever beam. The finite element method is used to build the model of the beam. The dynamic response data of the piezoelectric smart structures is obtained through the FEM analysis. According to the nonlinear approximation capability of neural networks, a dynamic recursive BP neural network is used to identify the piezoelectric vibration system, and the predictive model is built. On the basis of the neural network predictive model, the generalized predictive control algorithm is proposed to control the piezoelectric vibration system and to suppress the undesired vibration of the structures. Simulation results demonstrate the excellent performance of the developed control system. It has some significance for guiding the application of intelligent control algorithms in smart structures.