Cybersecurity In Smart Grids: Deep Learning Approaches To Intrusion Detection In Iot-Enabled Power Systems

Abstract

The increasing complexity and interconnectedness of IoT-enabled smart grids expose these systems to diverse cybersecurity threats, making effective intrusion detection essential. This research presents a comparative analysis of deep learning models—Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and Transformer models—for intrusion detection within smart grids. Each model was evaluated on performance metrics such as accuracy, precision, recall, F1-score, and computational efficiency. Results show that the Transformer model achieved the highest accuracy and F1-score, excelling in capturing complex temporal patterns critical for identifying sophisticated intrusions. The LSTM model demonstrated strong recall but required higher computational resources, while the CNN model, though computationally efficient, displayed limitations in capturing temporal dependencies.
The study also investigated the effect of hyperparameter tuning, with surface plot analyses revealing optimal learning rate and batch size combinations for each model to maximize detection accuracy. Radar and area graph analyses illustrated the models' performance across different metrics, highlighting CNNs for rapid detection, LSTMs for recall-intensive applications, and Transformers for high-accuracy scenarios. Compared to traditional machine learning methods, the deep learning approaches demonstrated superior accuracy, establishing them as viable solutions for enhanced cybersecurity in smart grids.
This work provides a foundation for deploying robust and scalable deep learning-based intrusion detection systems in IoT-driven power systems. Future research could further optimize these systems through hybrid or ensemble approaches to adapt to evolving cyber threats in smart grid environments.