An Explainable Deep Learning System for Cyberattack Detection in Internet of Energy Networks

The integration of smart devices and real-time analytics in the Internet of Energy (IoE) has enabled intelligent, data-driven energy systems but also introduced cybersecurity concerns. Addressing these concerns requires detection mechanisms that are not only accurate but also interpretable. This article proposes an explainable deep learning system, named E-SHIELD, designed to detect various cyberattacks in IoE networks. The system uses a hybrid Convolutional Neural Network—Long-Short-Term Memory (CNN-LSTM) architecture to learn both spatial correlations and temporal dependencies from network traffic data. The architecture is optimized for resource-constrained environments by using compact filters, dropout regularization, and low learning rates. E-SHIELD is evaluated using two benchmark datasets, WUSTL-IIoT-2021 and CICIoT-2023, which contain diverse real-world attack scenarios such as Denial of Service (DoS), Distributed DDoS, Mirai variants, and fuzzing attacks. To ensure transparency, SHAP (SHapley Additive exPlanations) is integrated to provide both global and local interpretability of the model’s predictions. A user dashboard has been developed to support human-in-the-loop monitoring, allowing users to analyze predictions, feature importance, and risk levels in real-time. Experimental results show that the E-SHIELD system achieves high accuracy compared to existing CNN, LSTM, and ensemble-based models. By combining robust detection capabilities with interpretable AI, E-SHIELD supports secure and trustworthy deployment in mission-critical energy systems.