Wei, Xiang, Zhou, Yue, Wang, Guibin, Hu, Ze, Zhu, Ziqing, Zhang, Xian, Chan, Ka Wing and Wu, Jianzhong  ORCID: https://orcid.org/0000-0001-7928-3602
2026.
Coordinated transmission-distribution load restoration under N-k contingencies: a distributed optimization and reinforcement learning approach.
Applied Energy
423
, 128328.
10.1016/j.apenergy.2026.128328
|
Preview |
PDF
- Accepted Post-Print Version
Available under License Creative Commons Attribution. Download (5MB) | Preview |
Abstract
Ensuring the rapid restoration of loads in transmission and distribution (T&D) systems under emergency conditions is crucial for maintaining grid operation. This study addresses the challenge of load restoration when contingencies, such as the disconnection of transmission lines and generators, disrupt the power supply. To address this issue, a coordinated T&D operation strategy is introduced in this work. VPPs within the distribution system are utilized to compensate for curtailed loads and support transmission-level load restoration. The coordination process involves bidirectional information exchange: the transmission system communicates load-shedding decisions to the distribution system, while the distribution system provides the available maximum curtailment capacity through VPPs. This interaction enhances the system's ability to respond to N-k contingency events in a distributed optimized manner, improving overall resilience. To achieve efficient decision-making in this coordinated framework, reinforcement learning techniques are employed to optimize load restoration under N-k contingencies. The transmission system is modeled using the soft actor-critic (SAC) algorithm, which determines optimal load-shedding and generator dispatch strategies for rapid system recovery. Meanwhile, the distribution system, responsible for managing multiple VPPs, is controlled using the complementary attention for the multi-agent SAC (CMS) algorithm. This approach mitigates the common attention dispersion problem in multi-agent SAC implementations, ensuring optimal decision-making in dynamic multi-agent environments. Simulation results demonstrate that the proposed reinforcement learning-based framework effectively reduces constraint violation in the transmission system while maintaining load supply and voltage stability in the distribution network.
| Item Type: | Article |
|---|---|
| Date Type: | Published Online |
| Status: | Published |
| Schools: | Schools > Engineering |
| Additional Information: | RRS policy applied |
| Publisher: | Elsevier |
| ISSN: | 0306-2619 |
| Date of First Compliant Deposit: | 11 August 2026 |
| Date of Acceptance: | 25 June 2026 |
| Last Modified: | 11 Aug 2026 09:15 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/188184 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions