An AI-Based Real-time Intrusion Detection System for Power Electronics-Dominated Grid: Attack on Inverters PQ Set-Points

This work presents a long short-term memory (LSTM) neural network based method for real-time detection of intruded power setpoints of grid-following inverters (GFLIs) in power electronics dominated grids (PEDG). The supervisory control layer in PEDG plays a crucial role in coordination and optimal dispatch of GFLIs. Thus, a compromised supervisory control layer and/or the communication line between the supervisory control and primary control layers can jeopardize the stable operation of the network. The attack vector considered in this paper manipulates active (P) and reactive (Q) powers setpoints. The proposed real-time intrusion detection system (IDS) is based on the effect of this attack vector on voltage, current, frequency and rate of change of frequency (ROCOF) sensed at inverters point of common coupling (PCC). Classifying the time series data into normal or anomalous types can help detect attacks on PQ setpoints in real-time. Various source and load disturbances along with anomaly/attack scenarios are considered for database creation and training the neural network. The proposed real-time IDS is then validated in the DER-based PEDG system. The neural network performance characteristics in terms of accuracy and loss proved it to be capable of effectively distinguishing anomalous behavior from regular fluctuations occurring in the PEDG.