Hybrid CNN BiLSTM architecture for smart grid cyberattack detection using smart meter data

This project proposes the implementation of a supervised deep learning framework designed for real-time cyber-attack detection within smart grid communication infrastructures. Addressing the limitations of traditional rule-driven methods against evolving intrusion tactics, the architecture integrates a Convolutional Neural Network (CNN) for extracting hierarchical spatial features from high-frequency, multi-modal smart meter measurements, such as voltage, current, and frequency harmonics. A Bidirectional Long Short-Term Memory (BiLSTM) component is subsequently employed to model temporal dependencies by processing sequences of meter data in both forward and backward directions, thereby capturing complex attack evolution patterns. Attention mechanisms are incorporated to dynamically weigh the relevance of temporal features, enhancing both prediction accuracy and model interpretability. For

robust decision-making, an Extra Trees ensemble classifier is utilized, providing a low-variance output as an alternative to standard Softmax layers. The framework is specifically engineered to overcome challenges such as class imbalance, high dimensionality, and noise inherent in sensor data from heterogeneous sources, aiming to deliver millisecond-level real-time intrusion detection capabilities.

Message: Hi FE, I need help with “Hybrid CNN BiLSTM architecture for smart grid cyberattack detection using smart meter data” in “Information Technology”