Deep Learning–Based Network Anomaly Detection in Cyber-Physical Systems Using Edge Hardware and Cloud Intelligence

Abstract

Modern network infrastructures have become highly vulnerable due to the rapid development of cyber-physical systems (CPS). This is because the constant creation of high-volume, high-velocity data streams requires real-time anomaly detection, which is highly computationally intensive and latency-sensitive. Although recent studies have explored deep learning and edge computing separately for anomaly detection, few studies have jointly integrated edge inference with cloud intelligence in a unified adaptive framework for CPS. Existing approaches generally suffer from one or more limitations, including high inference latency, high computational cost, insufficient scalability, or inability to continuously update deployed models. Therefore, there remains a need for an integrated architecture capable of providing accurate, scalable, and low-latency anomaly detection. In this paper, a solid anomaly detection framework is proposed that uses deep learning, integrating edge computing and cloud intelligence to enable efficient, scalable, and adaptable threat detection. The suggested system will include a structured pipeline that involves data acquisition, preprocessing, feature extraction, anomaly prediction, and adaptive feedback optimization. Low-latency inference can be achieved with edge nodes, while cloud resources can be used for computationally intensive training and model refinement. Experimental analysis shows that the suggested approach achieves 93.6% detection accuracy, 90.2% efficiency, 182 ms latency, a stability index of 0.89, and a total performance index of 0.91. The findings confirm the usefulness of the hybrid edge-cloud paradigm in enhancing the accuracy of detection, minimal response time, and high system resilience and CPS settings.