| Abstract |
This work develops a supervised learning-based strategy using Long Short-Term Memory Networks(LSTMN) to distribute power between fuel cells and batteries for fuel cell trains. The learning-basedstrategy exceeds the adaptive Pontryagin's minimum principle (PMP)-based strategy in all driving conditions, which is state-of-the-art. In the case with the most significant difference between the learning-based and the adaptive PMP-based strategy, more consumption of 1.84\% than the off-line PMP strategy, which determines the minimal hydrogen consumption for the same load profiles, is observed for the learning-based strategy. In comparison, 2.54\% more consumption is required by the adaptive PMP-based strategy compared to the off-line strategy. |
