A scalable, causal, adaptive rule-based energy management for fuel cell hybrid railway vehicles learned from results of dynamic programming

Abstract

Abstract A scalable, causal, adaptive rule-based energy management strategy for fuel cell hybrid trains is developed. The rules of this strategy are initiated by the results of two-dimensional dynamic programming under different driving conditions and utilize the convexity of the characteristic specific consumption curve of the fuel cell system. According to the strategy, the fuel cell power follows the estimated average load power. This average value is updated each time when the train leaves a station by using prior knowledge, which ensures its causality. Furthermore, the power demand due to the gradient slope is excluded while estimating the average value because the gravitational energy is recyclable. In this way, the fuel cell system works more stably without being influenced by the strongly changeable power demand due to the gradient slopes. In order to avoid over-charging of batteries during long hold time, which is often the case for regional railway vehicles, the pre-known driving, holding, and travel time available in railway transportation are used to improve the estimation of the average values. After comparison with the results of dynamic programming, an excellent fuel economy is observed under different driving cycles and weather. More consumption of 0.01% and 0.09% in summer and winter, respectively, compared to dynamic programming, results under a typical driving cycle of regional railway vehicles in Berlin. Because the rules are based on the component characteristics, this strategy can be transferred to other vehicle configurations or driving situations without a loss of effectiveness. In addition to the excellent fuel economy, the lifetime of fuel cell systems benefits from its less dynamic operation.