Toward Real-Time Energy Management: A Comparative Study of DP, PMP, and RL for AWD Electric Vehicles
IEEE Transactions on Intelligent Vehicles · 2026-11 · 11 (11) · 1261-1270
DOI: 10.1109/tiv.2026.3733892 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Intelligent Vehicles, 2026-11. Verfasst von Sang Rak Kim und 2 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Gefördert von: National Research Foundation of Korea, Korea government.
Abstract (Originalsprache)
Energy management in all-wheel-drive (AWD) electric vehicles is commonly formulated as an optimal control problem. Achieving real-time implementation while maintaining near-optimal performance remains a key challenge. This study presents a comparative analysis of Dynamic Programming (DP), a Pontryagin's Minimum Principle (PMP)-based method (SCEMP), and reinforcement learning (RL) under consistent modeling and simulation conditions. In particular, the differences in policy representations and their impact on optimality and real-time feasibility are systematically investigated. An RL framework based on a Markov Decision Process (MDP) is designed to integrate system observations with limited near-future velocity information. Simulation results show that the RL policy achieves near-optimal performance within 1% of the DP solution, while exhibiting control behavior similar to SCEMP. Furthermore, we deployed a lightweight RL model—optimized through knowledge distillation and quantization—on CPU- and GPU-based edge computing devices, confirming that the agent achieves control latency below 10 ms. These results demonstrate that RL provides a feasible and efficient solution for real-time energy management, while offering insight into the relationship between policy structure and control performance.
Förderung
- National Research Foundation of Korea
- Korea government
Zitieren
APA
Kim, S. R., Lee, M., & Cha, S. W. (2026). Toward Real-Time Energy Management: A Comparative Study of DP, PMP, and RL for AWD Electric Vehicles. IEEE Transactions on Intelligent Vehicles, 11(11), 1261-1270. https://doi.org/10.1109/tiv.2026.3733892
Vancouver
Kim SR, Lee M, Cha SW. Toward Real-Time Energy Management: A Comparative Study of DP, PMP, and RL for AWD Electric Vehicles. IEEE Transactions on Intelligent Vehicles. 2026;11(11):1261-1270. doi:10.1109/tiv.2026.3733892
BibTeX
@article{Kim2026,
author = {Kim, Sang Rak and Lee, Minji and Cha, Suk Won},
title = {Toward Real-Time Energy Management: A Comparative Study of DP, PMP, and RL for AWD Electric Vehicles},
journal = {IEEE Transactions on Intelligent Vehicles},
year = {2026},
volume = {11},
number = {11},
pages = {1261--1270},
doi = {10.1109/tiv.2026.3733892}
}
EUVIMED