Proper modeling of Thermal Management System (TMS) in Electric Vehicles (EVs) is crucial in terms of designing the EV components. Data-driven methods come up as an alternative to the computationally intensive high-fidelity methods or reduced order models where the accuracy is sacrificed for performance. In this paper, two informed neural network approaches are benchmarked in EV TMS modeling: Analytical Feature Engineering, where new features are generated by using the physical processes that take place within the EV, and Feature Generation via Loss Maps where loss maps of the inverter and the electric engine are used to generate a new power loss feature. Results show that accuracy increased by 1.7% to 3.6% depending on applied features and the neural network architecture.
«
Proper modeling of Thermal Management System (TMS) in Electric Vehicles (EVs) is crucial in terms of designing the EV components. Data-driven methods come up as an alternative to the computationally intensive high-fidelity methods or reduced order models where the accuracy is sacrificed for performance. In this paper, two informed neural network approaches are benchmarked in EV TMS modeling: Analytical Feature Engineering, where new features are generated by using the physical processes that tak...
»