Scholarlink

Data centers are rapidly becoming one of the largest consumers of electricity worldwide, with their share expected to rise from 2% to 8% of global usage by 2030. This paper introduces a federated learning (FL)–based predictive maintenance framework for green data centers, integrating LSTM-based failure prediction, differential privacy, and dynamic energy optimization. The system, deployed across five geographically distributed data centers over twelve months, achieved a 20.3% reduction in energy use, a 15.2% decrease in CO2 emissions, and a 95.09% F1-score in failure detection. The FL approach ensures privacy, scalability, and compliance, outperforming centralized and static-threshold baselines. The results demonstrate the potential of collaborative, privacy- preserving AI for sustainable digital infrastructure.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-Driven Predictive Maintenance for Green Data Centers: A Federated Learning Approach to Reduce Carbon Footprint

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shreesh Prateek Pathak,
  • Sivakumar Rajagopal

摘要

Data centers are rapidly becoming one of the largest consumers of electricity worldwide, with their share expected to rise from 2% to 8% of global usage by 2030. This paper introduces a federated learning (FL)–based predictive maintenance framework for green data centers, integrating LSTM-based failure prediction, differential privacy, and dynamic energy optimization. The system, deployed across five geographically distributed data centers over twelve months, achieved a 20.3% reduction in energy use, a 15.2% decrease in CO2 emissions, and a 95.09% F1-score in failure detection. The FL approach ensures privacy, scalability, and compliance, outperforming centralized and static-threshold baselines. The results demonstrate the potential of collaborative, privacy- preserving AI for sustainable digital infrastructure.