Intrusion Detection Using Time-Series Imaging and Transfer Learning in Smart Grid Environments

Intrusion detection systems (IDS) monitor and analyze network traffic and system activity to detect and alert security personnel to potential security breaches or attacks. Although deep learning models have shown great promise in improving the accuracy and efficiency of IDSs, several challenges are associated with their use, including data scarcity and model complexity. Furthermore, to overcome these problems, deep transfer learning is considered in this study. Typically, this article presents a novel intrusion detection (ID) approach using transformed 1D signals into 2D representations and applying pre-trained convolutional neural network (CNN) models. The transformed 2D representations of the signals allow the pre-trained CNN models to effectively learn the features of the signals and accurately classify them as normal or malicious. The performance of the proposed method was evaluated on the CIC-IDS-2018 dataset, and the results showed 92% accuracy in differentiating between normal behavior and malicious activities, which is an improvement compared with other detection methods.