Skip to main navigation Skip to search Skip to main content

Bi-level energy management for microgrid-integrated charging stations with TranSAC and non-cooperative game

  • Yihao Meng
  • , Yuan Zou
  • , Zhiying Zhao
  • , Guodong Du*
  • , Xudong Zhang
  • , Jun Zhang
  • *Corresponding author for this work
  • Beijing Institute of Technology
  • Beijing North Vehicle Group Corporation

Research output: Contribution to journal › Article › peer-review

Abstract

As critical facilities for transportation-energy coupling, microgrid-integrated charging stations (MGCSs) require online scheduling to enhance system economics and electric vehicle (EV) charging completion degree. However, existing deep reinforcement learning (DRL)-based studies assume ideal Markov Decision Process (MDP), failing to accommodate data privacy and EVs' non-cooperative behavior, which constitutes a Partially Observable Markov Decision Process (POMDP). This paper presents a bi-level economic scheduling framework to address this weakness. The upper-level coordinates energy storage and EV aggregation target power via a TranSAC algorithm, which integrates a feature-weighted Transformer with Soft Actor-Critic (SAC) to solve the POMDP and provide attention-based interpretability. The lower-level determinate vehicle-to-grid (V2G) power through a decentralized non-cooperative game, enabling feasible physical responses without with reduced exposure of private EV information. Simulations demonstrate that, the proposed framework achieves the closest performance to the theoretically optimal upper bound among implementable methods, with improvements in both economic benefit and EV charging completion degree.

Original languageEnglish
Article number142521
JournalEnergy
Volume367
DOIs
Publication statusPublished - 1 Dec 2026

Keywords

  • Microgrid-integrated charging station
  • Non-cooperative game
  • POMDP
  • Transformer

Fingerprint

Dive into the research topics of 'Bi-level energy management for microgrid-integrated charging stations with TranSAC and non-cooperative game'. Together they form a unique fingerprint.

Cite this