Energy Management Optimization for Connected Hybrid Electric Vehicle with Offline Reinforcement Learning

Niu Zegong, He Hongwen*, Wang Yong, Huang Ruchen

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

4 Citations (Scopus)

Abstract

Recently, offline deep reinforcement learning (DRL) which is a branch of DRL has been widely concerned and sets off an upsurge of researches related to it. In this paper, we apply a new algorithm called batch-constrained deep Q-learning (BCQ) to search the optimal control strategy for hybrid electric vehicles (HEVs). Then, thanks to the superiority of offline DRL, we put forward a scheduled training framework of energy management strategy (EMS) based on connected vehicles. The results show the BCQ based strategy performs well in both the stabilization of battery SoC and the fuel economy, which reaches 95.05% of DP. Offline DRL possesses a significant applied perspective and provides a new approach for power allocation of HEVs.

Original languageEnglish
Title of host publicationICEIEC 2022 - Proceedings of 2022 IEEE 12th International Conference on Electronics Information and Emergency Communication
EditorsLi Wenzheng
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages103-106
Number of pages4
ISBN (Electronic)9781665407540
DOIs
Publication statusPublished - 2022
Event12th IEEE International Conference on Electronics Information and Emergency Communication, ICEIEC 2022 - Beijing, China
Duration: 15 Jul 202217 Jul 2022

Publication series

NameICEIEC 2022 - Proceedings of 2022 IEEE 12th International Conference on Electronics Information and Emergency Communication

Conference

Conference12th IEEE International Conference on Electronics Information and Emergency Communication, ICEIEC 2022
Country/TerritoryChina
CityBeijing
Period15/07/2217/07/22

Keywords

  • Cloud Platform
  • Connected Vehicle
  • Energy Management Strategy
  • Offline Deep Reinforcement Learning

Fingerprint

Dive into the research topics of 'Energy Management Optimization for Connected Hybrid Electric Vehicle with Offline Reinforcement Learning'. Together they form a unique fingerprint.

Cite this