Streaming Data Anomaly Detection in Energy Big Data Center Using Unsupervised Learning
Aonan Wu, Chengyuan Zhu, Xuejun Jiang, Qinmin Yang · 2025
The energy big data center plays a pivotal role in supporting real-time monitoring and decision-making. The complexity and dynamic nature of data streams makes anomaly detection a critical task in such systems. This study proposes an unsupervised learning framework, Unsupervised Learning-based Streaming Anomaly Detection (UL-SAD), for time-series anomaly detection in streaming data from the energy big data center. The framework integrates modules for data preprocessing, fluctuation analysis, dual-phase smoothing, and automated thresholding based on the Generalized Pareto Distribution (GPD). It achieves efficient anomaly detection in high-dimensional, noisy environments without relying on labeled datasets. Experimental results demonstrate that UL-SAD achieves an anomaly detection accuracy of 88% on multidimensional energy datasets, significantly improving operational efficiency and accuracy.