This paper presents a comparative study of two promising real-time energy management strategies for fuel cell electric vehicle applications: adaptive equivalent consumption minimization strategy (A-ECMS), and stochastic dynamic programming (SDP). An off-line algorithm –classic dynamic programming - provides reference results. On-line and off-line strategies are tested both in simulation and using a dedicated test bench completely consistent with an electric scooter powertrain. The hybrid power source combines a fuel cell, a supercapacitor pack and two related power converters. The system model is carefully calibrated using experimental data. This allows meaningful identification of parameters of the various strategies. The model data is determined using a motorcycle certification driving cycle. The robustness of each strategy is then analyzed using a large number of random driving cycles. Experimental and simulation results show that a specific SDP approach, based on Markov chain modeling, has the best overall performance in real-world driving conditions. It achieves the minimum average hydrogen consumption while respecting the state-sustaining constraint. Conversely, the A-ECMS results lack robustness and show poor performance indexes when facing unknown real world power demand profile. In conclusion, the present results indicate SDP is an interesting approach for future hybrid source energy allocation.