Keywords
server power anomaly detection
noisy time series
diffusion denoising
disentangled representation
workload fluctuation
robust anomaly detection
Abstract
Data centers continuously collect power-related time series from server power draw, rack-level energy consumption, CPU utilization, fan speed, inlet temperature, power distribution units, workload scheduling, and cooling interaction signals. These signals are noisy because of workload bursts, virtual machine migration, cooling feedback, transient load balancing, sensor fluctuation, and power measurement jitter. Raw power anomalies are difficult to interpret because normal workload-induced changes may resemble early hardware degradation or abnormal energy use. This study proposes a disentangled power signal denoising method for robust server anomaly detection in data centers. The method applies a conditional diffusion model to reconstruct stable power-consumption trajectories from noisy measurements. A disentanglement module separates workload-driven variation, cooling-related fluctuation, and server-side abnormal residuals. Anomaly scores are computed from denoised power residuals and rack-level consistency constraints. Experiments are conducted on a data center power monitoring dataset containing 18 server rooms, 9,800 physical servers, 620 racks, 74 power distribution units, and 46 monitoring variables collected at 15-second intervals over 11 months. The dataset contains 1.92 billion timestamped records and 3,180 operations-confirmed abnormal episodes, including abnormal power spikes, fan-control instability, rack-level energy imbalance, degraded power supply units, unexpected idle power increase, and cooling-power coupling faults. The proposed method reduces median detection delay from 26.4 minutes to 7.5 minutes compared with a temporal denoising autoencoder. False operation tickets decrease to 1.8 cases per server room per week. Diffusion denoising lowers normalized power reconstruction deviation from 0.141 to 0.052, and workload-disentanglement removes 68 million burst-load windows from the abnormal candidate queue. The online scoring engine processes 72,000 power records per second with 38 ms median inference latency per rack window. The results demonstrate that diffusion-based denoising and disentangled representation learning can improve robust anomaly detection in noisy data center power time series.
References
1. Fu, Y., Gui, H., Li, W., & Wang, Z. (2020, August). Virtual Material Modeling and Vibration Reduction Design of Electron Beam Imaging System. In 2020 IEEE International Conference on Advances in Electrical Engineering and Computer Applications (AEECA) (pp. 1063-1070). IEEE.
2. Lee, J., Yoon, T., Koo, W., & Kim, H. (2026). Knowledge-Assisted Multi-Graph Dependency Learning for Multivariate Time Series Anomaly Detection in Multi-Stage Industrial Processes. IEEE Transactions on Automation Science and Engineering.
3. Chen, X., Xiao, H., Zeng, Z., Zhang, S., & Xiao, R. (2025). Fine-Grained Multivariate Time Series Anomaly Detection via Causal Inference. Knowledge-Based Systems, 114765.
4. Sharifi, S., Stocco, A., & Briand, L. C. (2025). System Safety Monitoring of Learned Components Using Temporal Metric Forecasting. ACM Transactions on Software Engineering and Methodology, 34(6), 1-43.
5. Liang, R., Feifan, F. N. U., Liang, Y., & Ye, Z. (2025). Emotion-Aware Interface Adaptation in Mobile Applications Based on Color Psychology and Multimodal User State Recognition. Frontiers in Artificial Intelligence Research, 2(1), 51-57.
6. Wu, C., Zhang, F., Chen, H., & Zhu, J. (2025). Design and optimization of low power persistent logging system based on embedded Linux.
7. Habiba, M., Pearlmutter, B. A., & Maleki, M. (2026). Different Neural Network Models for Time Series Processing. In Recent Trends in Modelling the Continuous Time Series Using Deep Learning (pp. 27-61). Cham: Springer Nature Switzerland.
8. Shao, W. (2026, March). Interpretable Ensemble Learning for Network Traffic Anomaly Detection: A SHAP-based Explainable AI Framework for Embedded Systems Security. In 2026 14th International Symposium on Digital Forensics and Security (ISDFS) (pp. 1-4). IEEE.
9. Gao, G., Gao, R., Lu, C., Gao, R., & Kuang, Y. (2026, March). Security Governance Methods and Quantitative Evaluation for Enterprise SMS and Verification Code Systems. In 2026 International Conference on Generative Artificial Intelligence and Information Security (GAIIS) (pp. 455-458). IEEE.
10. Attila, K., Molnár, J. K., & Károly, J. (2026). Correction Functions and Refinement Algorithms for Enhancing the Performance of Machine Learning Models. Automation, 7(2), 45.
11. Zhang, Z., Gao, Y., & Tong, Y. (2026). CausalML4CX: A Causal Inference and Machine Learning Framework for Customer Experience Optimization with Application in Banking.
12. Xu, T., Zhu, W., & Zhang, J. (2026). A Study on the Application of Alternative Data in Credit Assessment for the Unbanked Population.
13. Gyang, P. P., Chakraborty, P., Meegahapola, L., & Yu, X. (2026). Grid Integrated Data Centers: Architectures, Modeling, Interconnection Requirements, Challenges, and Opportunities. IEEE Open Journal of the Industrial Electronics Society.
14. Xu, T., Zhang, J., & Zhu, W. (2026). Fairness-Accuracy Trade-offs and Calibration Mechanisms in Machine Learning-Based Subprime Credit Scoring. Available at SSRN 6893759.
15. Wang, Y., Chen, J., Arias, R., Wang, Y., & Yin, X. (2026). Development and Validation of a Patient-Friendly Digital Assessment Platform for Precision Screening of Oral Anti-Obesity Medications (AOMs).
16. Rawat, R., Borana, K., Chaturvedi, N., Rawat, H., Rawat, A., Rajavat, A., & Rimal, Y. (2026). Deep Learning and Pattern Recognition for Anomaly Detection in Multimodal Medical Imaging With Sensor‐Based Data Fusion. Journal of Sensors, 2026(1), 9935004.
17. Zhu, W., & Yang, J. (2025). Causal Assessment of Cross-Border Project Risk Governance and Financial Compliance: A Hierarchical Panel and Survival Analysis Approach Based on H Company's Overseas Projects.
18. Jiao, Y., Zhao, B., Wang, A., & Shi, T. (2026). Construction and Empirical Study of a Modularized Teaching System for Art Courses Based on a Unified Training Pathway.
19. Erkek, T. Ü., Erkek, M., Aydın, M. B., & Gezer, K. (2025). Experimental performance characterization of R-454B and R-410A in data center server rack mount cooling unit: Energy efficiency and environmental impact assessment. International Journal of Refrigeration.
20. Yin, J., Huang, Y., & Rao, H. (2026). A Study on the Dynamic Evolution of Learning Behavior and Outcome Prediction in Digital Educational Environments. Available at SSRN 6607139.
21. Yang, J. (2026). Computational Analysis of How Digital Non-Clinical Communication Tools Influence Social Participation Among Older Adults in Community-Based Elderly Care. Available at SSRN 6682838.
22. Hinov, N. (2025). From Energy Efficiency to Energy Intelligence: Power Electronics as the Cognitive Layer of the Energy Transition. Electronics, 14(23), 4673.
23. Zhang, Z. (2026). A Study on Multimodal Product Understanding and Assembly Guidance Optimization in e-commerce for Complex Consumer Goods. Available at SSRN 6734559.
24. Qi, C., & Qiao, X. (2026). Building and Operating a Large Scale Multi-Agent System: A Case Study from Industry. Available at SSRN 6795198.
25. Khatun, Z. (2025). AI-Driven Predictive Maintenance For Motor Drives In Smart Manufacturing A Scada-To-Edge Deployment Study. American Journal of Interdisciplinary Studies, 6(1), 394-444.
26. Su, D., & Wu, Y. (2026). Multilevel Growth and Social Network Modeling for Functional Communication Enhancement in Students with Intellectual Disabilities.
27. Moshawrab, M., Adda, M., Bouzouane, A., Ibrahim, H., & Raad, A. (2023). Reviewing Federated Learning Aggregation Algorithms; Strategies, Contributions, Limitations and Future Perspectives. Electronics (2079-9292), 12(10).
28. Xie, Z., Ren, X., Zheng, T., Bai, J., Fan, W., Xu, B., ... & Song, Y. (2026). A Survey on AI Agent Harness. ResearchGate Preprint. DOI, 10.
29. Naji, H. R., & Riyahi, H. (2025). A Survey of Load Balancing Approaches Based on Switch Migration in Software‐Defined Networks. IET Networks, 14(1), NA-NA.