A Feedforward Power Coordination Control Strategy for Hydrogen Fuel Cell Hybrid Trains Considering Dynamic Response Delay

Hydrogen fuel cell hybrid trains are normally sized so that the fuel cell supplies the average or base power demand, while the traction battery accommodates short-duration power transients and regenerative braking. Even when the adopted steady-state power ratings satisfy the specified operating duties, however, the finite fuel-cell ramp rate and command-to-power delay can create brief power mismatches during abrupt operating-mode transitions. This paper proposes a route-informed, prediction-assisted feedforward power coordination strategy. A Transformer predictor estimates the 10 s traction-power demand using route information and simulated operating-state data, and the predicted sequence is supplied to a constrained model predictive control (MPC) controller as preview information. A reduced-order first-order fuel-cell response model is used to characterize the dominant command-to-power lag and determine the advance-command timing at the train energy-management level, while the unit-level ramp-rate limits are retained as implementation constraints. An offline normalized Pareto sweep is further conducted to determine the MPC weighting coefficients for hydrogen consumption, fuel-cell power fluctuation, and state-of-charge (SOC) deviation. The present simulations quantitatively evaluate the traction-power prediction and delay-aware advance-command mechanisms within the proposed constrained MPC framework. For the investigated flat-track operating cycle, the predictor achieves a root mean square error (RMSE) of 414.93 kW, a mean absolute error (MAE) of 140.65 kW, and a coefficient of determination (R2) of 0.9883 on held-out samples from the same operating distribution. Among 231 feasible supervisory-layer candidate weight combinations, 134 non-dominated solutions are retained, and the knee-point criterion yields w1=0.50, w2=0.50 and w3=0.00. During a representative transition to regenerative braking, the advance command reduces the pre-saturation battery charging-power request from 3.7 MW to 3.0 MW. Relative to the adopted aggregate converter-side charging limit, the corresponding peak residual braking power decreases from 1.3 MW to 0.6 MW. The contribution lies in explicitly coordinating route information, short-horizon power prediction, fuel-cell response delay, and constrained power allocation. The present results are simulation-based; nonlinear balance-of-plant dynamics, battery degradation, cross-route generalization, and train control and management system (TCMS) hardware-in-the-loop timing remain to be validated.