The commercial development of proton exchange membrane fuel cells (PEMFCs) in automotive applications is constrained by lifespan issues. Lifespan prediction research is essential for enhancing reliability by improving parameter control. Most existing studies predict PEMFC lifespan under steady-state conditions, yet real-world scenarios involve dynamic operating conditions with frequent startups, shutdowns, and power fluctuations. Therefore, there is a growing focus on predicting PEMFC performance under dynamic conditions. This study simulates actual vehicular dynamics through periodic cycle composite operating experiments to examine PEMFC characteristics. A PEMFC health indicator is established using mean voltage at 85% rated current output per cycle. The effectiveness of this indicator is verified by comparing it with voltage decay characteristics from polarization curves. Subsequently, to address the need for long-term lifespan prediction under composite operating conditions, a PEMFC lifespan prediction model is developed using Elman neural networks, with model parameters optimized using genetic algorithms. Experimental results show that the optimized Elman neural network lifespan prediction model, based on the newly established health indicator, achieves high accuracy with a root mean square error of 0.007. Compared with various existing models and algorithms, it is simpler and more accurate.