An interpretable deep learning approach for early detection, classification and mitigation of cyberattacks in cyber-physical microgrids

The integration of Electric Vehicle Charging Stations (EVCSs) within microgrid systems has introduced new cybersecurity challenges. These infrastructures are increasingly targeted by cyberattacks that can disrupt the grid's operation. This paper presents a new holistic approach that enables early detection, accurate classification, and effective mitigation against these attacks. The proposed approach employs the Continuous Wavelet Transform (CWT) on the power demand data to generate time–frequency scalograms that effectively represent the attack features. The morphological image processing technique is then used to enhance the extraction of the attacks’ features from the generated scalogram images. The enhanced scalograms are then fed to the Convolutional Neural Network (CNN) to learn the relevant features from these scalograms and classify the types of cyberattacks. Furthermore, the Long Short-Term Memory (LSTM) network is incorporated to provide interpretability and valuable insights into the attack patterns and emerging threats. Moreover, a new dynamic sequential home charging mechanism is introduced to proactively mitigate the impacts of cyberattacks on the microgrid’s assets. The results have shown that the proposed approach was able to achieve a high detection and classification accuracy of 99.07% and 98.15%, respectively, while effectively mitigating the impact of cyberattacks and ensuring that all microgrid assets operated within their normal limits.