Securing Grid-Connected Packed E-Cell Multilevel Inverter: A LSTM-AE Approach to Hybrid Attack Mitigation

Abstract

The deployment of open communication infrastructure into power systems has drawn much attention due to its significant benefits, such as real-time monitoring, diagnostics, and regulatory purposes. But utilizing such technologies poses security challenges to the cyber-physical power systems (CPPS), which can highly degrade their operation. In this paper, a defense mechanism is adopted to tackle the hybrid attacks, including the Denial-of-Service (DoS) and false data injection (FDI) attacks in the grid-connected multilevel inverters from a systematic point of view. The proposed protection mechanism for the grid-connected multilevel inverter is realized in two stages. $(i)$ A Long-Short Term Memory based on autoencoder (LSTM-AE) is developed to detect DoS attacks, and an event-trigger mechanism based on Lyapunov theory is implemented to eliminate the effect of false data. $(ii)$ A sliding mode observer is adopted to recognize FDI threats, where the false data is eliminated by injecting the negative value of the identified false data. A prototype of a grid-connected nine-level Packed E-Cell (PEC9) topology as a targeted multilevel inverter is constructed to experimentally validate the feasibility of the proposed cyber resilience scheme for CPPS in microgrid applications. Note to Practitioners—The motivation of this work comes from the issue that gird-connected multilevel inverters, which can realize the large-scale application of sustainable generation units, are susceptible to cyber threats. The cyber threats will introduce security problems to the cyber-physical power systems (CPPS) with the photovoltaic (PV) modules. False data injection (FDI) and denial-of-service (DoS) attacks, as the most common cyber-attacks, can seriously compromise CPPSs’ performance. In this regard, the current work develops a two-stage defense mechanism to tackle cyber-attacks. The proposed cyber resilient framework is designed to identify and mitigate hybrid attacks, including DoS and FDI attacks. In particular, the Long-Short Term Memory based on autoencoder (LSTM-AE) is utilized to identify the DoS attack, while the FDI attack is detected by the sliding mode observer (SMO). An event-triggered mechanism is also developed in the proposed defense algorithm to block the signal falsified by the DoS attack and submit the signal predicted by LSTM-AE to the system’s controller. The SMO estimates the FDI disruption to the system and injects it into the system measurement signal to eliminate the FDI attack’s impact on the system dynamic performance. For this approach, the critical challenge is to develop the proposed protection mechanism in practical applications. To address this difficulty, the experimental examinations are carried out by building a prototype of the PEC9 inverter.