NRASecure: A deep learning-based framework for enhancing security in 5G-enabled smart grid networks

Abstract

The integration of 5G networks with smart grid infrastructure introduces critical cybersecurity vulnerabilities. This requires advanced anomaly-detection frameworks capable of real-time threat identification. Traditional deep learning approaches suffer from insufficient spatial invariance, dependence on large training datasets, and reduced effectiveness against subtle cyberattacks targeting smart grid components. This paper presents NRASecure, a novel deep learning framework that combines canonical correlation analysis for feature selection with Jaspen's correlative convolutional neural network to enhance security in 5G-enabled smart grid networks. Canonical correlation analysis optimizes feature selection by computing cross-covariance matrices between two feature sets and identifying canonical vectors that capture the strongest linear relationships for cyberattack detection. The NSL-KDD dataset demonstrates the framework's effectiveness across varying dataset configurations, achieving a specificity of 97.61%, accuracy of 98.26%, precision of 99.40%, recall of 96.83%, F1-score of 98.10%, and memory consumption of 3.37 MB. In an additional experiment, NRASecure was also validated on the WUSTL-IIoT and 5G-NIDD datasets.