Abstract:PEMFC is a complex system which is often affected by many parameters. Its time-varying performance, nonlinear performance and uncertainty lead to its dynamic output unable to be controlled. Taking that into account, the controller based on adaptive prediction control was designed to control the dynamic output of PEMFC. The neural network model of PEMFC system was used as a predictive model of the controller, which could make full use of the nonlinear fitting ability of neural networks. Through optimizing the neural network connection weights and threshold in real time, receding horizon optimization could be realized, which could guarantee the real-time control and the effective electrochemical reaction. During the control process, the control precision and response speed of the system could be improved by the introduction of the negative feedback control to the front of adaptive model and predictive control. The simulation results show that the adaptive prediction controller owns adaptive ability, stronger robustness, stronger learning ability and higher control precision, and it has perfect control effect on the dynamic stability of proton exchange membrane fuel cell output characteristics.