Hardware-in-the-loop evaluation demonstrates a 24.3% cost reduction in four-wheel drive hybrid electric vehicles, indicating viable real-time adaptive power management.
Key Points
To develop and evaluate an adaptive, physics-informed digital twin energy management framework for four-wheel drive fuel cell hybrid electric vehicles utilizing dynamic wireless charging.
Integrated physics-informed neural networks (PINNs), soft actor-critic (SAC) deep reinforcement learning, and model predictive control (MPC) with an Elastic Weight Consolidation digital twin updating every 50 cycles.
Engineered real-time five-degree-of-freedom misalignment compensation for dynamic wireless power transfer and degradation-aware vehicle-to-grid (V2G) power dispatch.
Tested the framework across 200+ hours of hardware-in-the-loop (HIL) simulation using a dSPACE and NVIDIA Jetson platform under dynamic lateral misalignment with 50 mm displacement amplitude.
Achieved a 24.3% overall cost reduction and reduced battery degradation by 31.8% while generating €582.50 per year in vehicle-to-grid revenue.
Attained a mean wireless power transfer efficiency of 91.5% under dynamic lateral misalignment, delivering a 5.57-year discounted payback period and a 10-year net present value of approximately €3777.
Satisfied real-time operational constraints within a 100-ms limit, with PINN battery state estimation averaging 1.1 ms (worst-case 2.8 ms) and MPC execution averaging 32.1 ms (worst-case 62.4 ms).