A bistable polyvinylidene fluoride （PVDF） beam piezoelectric vibration energy harvester is designed. And its structural characteristics and working principle are introduced. In order to solve the deviation between the traditional theoretical model prediction and the actual output performance of the energy harvester， the artificial neural network is used to model the nonlinear relationship among its structural parameters， excitation frequency and electric energy. In this paper， an artificial neural network model of the bistable PVDF beam piezoelectric energy harvester is established based on a multi-layer feedforward network trained by error back propagation. The mass of the mass block， the compression distance of PVDF piezoelectric beam and the frequency of external excitation force are taken as input variables， and the output voltage root mean square （RMS） value of the harvester is taken as output variable. The experimental data of the piezoelectric energy harvester under different conditions are collected. The validity of the model is proved by comparing the simulation results. It is shown that the developed artificial neural network can effectively predict the output characteristics of the piezoelectric energy harvester without the need for complex theoretical modeling of the harvester.