Abstract
Deep learning techniques have been widely adopted in resilient control for distributed microgrid (MG). However, in practical implementations, due to data privacy concerns and limited computational resources, it is often impractical to transmit all raw data to a centralized server for model training. This paper investigates the distributed resilient secondary control problem for hybrid AC/DC MGs under hybrid false data injection (FDI) and denial-of-service (DoS) attacks with constrained resource budgets. Specifically, an adaptive federated learning mechanism is proposed, which dynamically adjusts the number of local training steps in each global aggregation round according to the available resource budget. Additionally, a reference tracking scheme is designed to estimate aggregated communication signals for resilient control, with cyber-attack mitigation and system stability rigorously guaranteed via Lyapunov-based analysis. Compared to existing data-driven resilient control methods, the proposed approach eliminates the need for distributed nodes to transmit raw data to a centralized location, and achieves enhanced prediction performance even under constrained resources. Finally, the effectiveness of the proposed control strategy against hybrid cyber-attacks is validated through a real-time experiment conducted on the OPAL-RT platform.
Funding Information
  • National Natural Science Foundation of China (62303241)
  • Scientific Starting Fund from Nanjing University of Posts and Telecommunications (NY223174)

This publication has 28 references indexed in Scilit: