A New Hybrid Wavelet Transform-Deep Learning for Smart Resilient Inverters in Microgrids Against Cyberattacks

Smart inverters are vital to the functioning and stability of contemporary microgrids (MG). Nevertheless, their growing dependence on digital communication and control exposes them to cyberattacks, which can disrupt grid stability and lead to equipment damage or outages. False Data Injection Attacks (FDIA), a severe cyberattack that intentionally falsifies sensor data, are particularly dangerous as they can mislead control systems. This paper proposes a new hybrid deep learning framework that integrates Long Short-Term Memory (LSTM) networks with the Gaussian Continuous Wavelet Transform (GCWT), a signal processing technique excelling at time-frequency feature extraction from power signals to detect severe FDIA in smart inverters, which are designed to significantly alter operational data and destabilize the grid. Comparative experiments with other wavelet transforms and deep learning architectures, such as Convolutional Neural Networks (CNN), Morlet continuous wavelet transform (MCWT), and Haar discrete wavelet transform (HDWT)-based models, validate the better performance of the proposed LSTM-GCWT model. It achieved 99.9% accuracy, surpassing all the baseline and hybrid methods. Moreover, a prediction-level comparison illustrates the strength of the model with 99% classification accuracy in identifying specific FDIA scenarios. The findings confirm the effectiveness of the LSTM-GCWT model in enhancing the security and reliability of smart inverters against FDIAs in MGs. While the model shows high performance, further research is needed to validate its generalizability across different inverter hardware and against novel, zero-day attack variants.