Predictive energy management strategy for fuel cell vehicles to mitigate the adverse effects of cumulative and propagated vehicle speed prediction errors

Vehicle speed prediction can improve the energy management strategy for fuel cell vehicles by anticipating future load changes. However, the cumulative and propagated amplification of vehicle speed prediction errors increases their impact on the performance of the strategy. To address this issue, this paper proposes a predictive energy management strategy based on enhanced random exploration and the fusion of internal and external multi-state predictions. Specifically, to mitigate training instability caused by vehicle speed prediction errors and enhance random exploration capabilities, a twin delayed deep deterministic policy gradient agent model with a variance-aware mechanism is constructed. From a data-driven perspective, to prevent external vehicle prediction errors from subjecting the fuel cell and power battery to extreme operating conditions, the future time-domain trends of the internal state of charge are considered. Using the predicted state variables as inputs for the agent, an energy management strategy is formulated that integrates multi-state predictions from both internal and external sources. A hardware-in-the-loop platform is established to evaluate the proposed method under both a standard driving cycle and a real-road random driving cycle. Compared with the baseline strategy, the proposed method reduces equivalent hydrogen consumption by 9.48% under the standard driving cycle and by 5.41% under the random driving cycle, thereby improving the overall performance of the strategy.