Cyber–physical anomaly detection for inverter-based microgrid using autoencoder neural network

Abstract

Contemporary cyber–physical systems have evolved into highly autonomous and distributed entities, enabled by cutting-edge control frameworks and advanced communication networks. This transformation has empowered these systems to efficiently accommodate inverter-based resources, harnessing the grid’s dynamic potential. However, the increased integration of cyber and communication infrastructure has exposed these systems to vulnerabilities stemming from both cyber and physical disturbances. Power system errors in the physical layer of microgrids and malicious attacks on the communication layer pose substantial threats to grid stability and load supply. In response to these challenges, this paper presents an innovative unsupervised learning method for cyber–physical anomaly identification. Leveraging an autoencoder-based neural network, this approach addresses the intricate interplay between the physical and cyber aspects of power system failures and false data injection within the communication network. The method’s efficacy is evaluated on an islanded inverter-based microgrid, representing a realistic and vulnerable testbed. The autoencoder error is introduced as a key performance metric, providing a robust and practical measure of the proposed detection technique’s effectiveness. Simulation results demonstrate the promising utility of this method, positioning it as a valuable tool for enhancing the resilience of contemporary cyber–physical systems against cyber–physical anomalies.