Localization of False Data Injection Attacks in Smart Grids: A Deep Learning Framework Based on SAE-CNN

The widespread adoption of smart grids has greatly enhanced grid operation and management through automation, real-time monitoring, and data collection capabilities. However, the open and distributed architecture of smart grids exposes them to various cyber threats, particularly False Data Injection Attacks (FDIA). FDIAs manipulate sensor data to disrupt grid operations, potentially causing erroneous state estimations and flawed scheduling decisions, which can compromise grid security and stability. To address this challenge, this paper introduces SAE-CNN, a hybrid method combining Sparse Autoencoder (SAE) and Convolutional Neural Network (CNN) for effective detection and localization of false data injection attacks. The approach begins with a sparse autoencoder for feature learning, extracting anomalous patterns from grid operation data to enable efficient compression and anomaly detection. The extracted features are then fed into a convolutional neural network, leveraging its strong spatial pattern recognition capabilities to accurately pinpoint the sources of false data injections. Experimental results demonstrate that the SAE-CNN method achieves high accuracy and robustness in identifying attack nodes, significantly enhancing the smart grid's defense against false data injection attacks.