Abstract: For the remaining lifetime prediction problem of proton exchange membrane fuel cell (PEMFC), this paper proposes a hybrid prediction method combining data-driven and model-driven by direct prediction approach based on a dynamic semi-empirical model of PEMFC considering the double-layer capacitance effect. For the data-driven approach, features of multi-dimensional aging data are extracted using a deep convolutional network and passed to a long and short-term memory network for aging voltage prediction. For model-driven, the voltage predictions are used as observations in an adaptive extended Kalman filtering framework. Short-term and long-term predictions are performed using hybrid prediction methods based on aging data under two operating conditions, static and dynamic, respectively. The short-term prediction results show that the dynamic semi-empirical model can fit the aging voltage data more effectively under dynamic conditions. The long-term prediction results show that the prediction error based on the dynamic semi-empirical model is smaller, and the remaining useful life of PEMFC predicted by the hybrid method is closer to the real value.