Anomaly Detection on Power Consumption of Network Devices

With the growing emphasis on Environmental, Social, and Governance (ESG) principles, reducing the power consumption of network devices has become a critical objective for Communication Service Providers (CSPs). This paper presents a machine learning-based anomaly detection method designed to identify abnormal power consumption in network devices, which may indicate hardware malfunctions or inefficiencies. The proposed method integrates an AutoEncoder (AE) model with historical power consumption analysis to detect anomalies. The AE is trained on real-world data collected from Chunghwa Telecom’s operational network and identifies deviations based on reconstruction errors. The method was implemented in a live network environment to monitor Optical Line Terminal (OLT) devices. The result showed it effectively identified subtle anomalies that traditional methods might overlook. This approach supports ESG goals by improving energy efficiency and operational reliability in network infrastructures.