Physics-guided thermal-aware eco-driving for heavy-duty electric trucks: A hierarchical MPC approach

Journal article, 2026

Heavy-duty electric trucks can reduce freight-transport emissions, but high payloads, prolonged uphill operation, and sustained traction demand couple energy-efficient driving with traction-motor thermal safety. Existing eco-driving studies rely on fixed efficiency maps or simplified motor-power models, whereas thermal-management studies often treat driving demand as an external input. This separation limits the controller’s ability to anticipate how speed planning alters temperature-dependent motor loss and the cooling demand. This paper proposes a physics-guided thermal-aware hierarchical model predictive control (MPC) framework for heavy-duty electric trucks. A differentiable semi-physical motor-loss surrogate is identified from multi-temperature loss data generated through a GT-SUITE/FEMAG workflow based on two-dimensional electromagnetic finite-element simulations. The surrogate maps motor speed, torque, and winding temperature to heat generation and achieves an RMSE of 1.106 kW and an R² of 0.9753 against the fitted GT-SUITE/FEMAG data set. The upper layer performs spatial-domain speed planning subject to winding-temperature constraints, while the lower layer schedules pump and fan commands in the time domain using the predicted traction demand as a thermal-load forecast. On the studied route, thermal-constrained speed planning reduces drive-system energy by 22.1% relative to cruise control and limits the maximum winding temperature to 149.32 (Formula presented). Relative to thermal-constrained speed planning with linear cooling, the hierarchical MPC reduces total energy from 24.151 to 23.455 kWh and cooling energy by 23.8%, while maintaining the thermal constraint. Together with controller-HIL evidence of real-time closed-loop execution, these results show that embedding temperature-dependent motor loss in eco-driving can deliver coordinated energy saving and traction-motor thermal protection within a hierarchical predictive-control framework.