Authors

T. W. Yoon, M. S. Song, H. Lee, Hyundai Motor Company, Hwaseong, Korea

Year

2026

Print Info

Production/Publication ÖVK

Summary

The global automotive industry is rapidly transitioning toward xEVs. This paper proposes a reinforcement-learning-based calibration method to improve the energy efficiency of a Hybrid Electric Vehicle (HEV) equipped with the TMED-II HEV drive system by calibrating the engine on/off boundary (EV line) and the engine operating point. A physics-based HEV simulator was developed by modeling the vehicle hardware and the control software, including the hybrid control unit (HCU) and transmission control unit (TCU), and was used as the reinforcement learning environment. The EV line was simplified into an analytical expression with two inputs (vehicle speed and SOC) and three calibration coefficients (α, β, γ). For fair fuel-consumption comparison, a target SOC was defined at the end of the driving cycle and SOC-Balancing was enforced. The SOC-Balancing coefficient γ was derived using the SCALE (SOC-Convergent Adaptive Learning Estimator) algorithm, which iteratively predicts γ, validates the resulting SOC through simulation, and updates the model until the target SOC is satisfied within a specified tolerance. In addition, an SGD-based reinforcement learning approach was applied to identify the optimal equivalence factor λ and coefficient β. In this procedure, fuel-consumption results obtained at randomly sampled λ values were interpolated to construct f(λ), and λ was iteratively updated using numerically estimated gradients to locate a minimum-fuel point. The derived optimal calibration was implemented in a test vehicle controller and validated on a chassis dynamometer with road-load settings. Under FTP and HWFET driving cycles, SOC matching to the target SOC reached 99.1% and the average fuel-economy improvement was 2.05%, demonstrating that the proposed calibration method is effective beyond simulation at the test-vehicle level.

ISBN

978-3-9504969-5-6

DOI

https://doi.org/10.62626/1oie-d4in Copy link

Order Members Login