Cyber-Attack Detection Approach In Photovoltaic Microgrids Using LSTM-based Autoencoders

Photovoltaic (PV) microgrids have been exposed to serious cyber threats for the integration of information and communication technologies (ICT) and renewable energy. In this article, a data-driven detection approach using an LSTM-based autoencoder has been developed against cyberattacks including fake data injection attacks (FDIA) and denial of service (DoS) attacks in PV microgrids. The proposed detection algorithm applies the unsupervised learning method that uses an LSTM neural network to capture the temporal feature and autoencoder to get reconstructed residuals of given input sequence data of measurements. By comparing residuals with the given threshold, the decision can be made on whether the system is attacked. The proposed algorithm has been evaluated for its performance in the PV microgrid model considering PV generation, energy storage system, and grid-tied function under various cyberattacks.