Comparative Study of Artificial Intelligence based Energy Management Strategies for Hydrogen Railway Vehicles

This work presents a comparative study between learning-based energy management strategies oriented to hydrogen railway vehicles, where a cost-optimal power split between the fuel cell and the battery is aimed. The proposed strategies aim to replicate the optimal operation solved by the dynamic programming optimization, which cannot be directly implemented in real-time. Specifically, three artificial intelligence-based strategies are proposed (ANFIS, Regression Trees and Ensemble Learning), and their results are compared against a baseline strategy and the aimed dynamic programming. Results are obtained for four driving cycles coming from two railway lines, and it is demonstrated that Ensemble Learning obtains the best operation among the proposed learning-based strategies. Indeed, compared to the optimal result of dynamic programming, the total cost of ownership of Ensemble Learning is 1.8-2.2% higher in Line A, and 6.3-8.0% higher in Line B. However, it is also found that in some of the proposed scenarios the baseline strategy obtains a better result, due to the fact that the optimal operation itself is close to a constant fuel cell operation.