A green data center cold load prediction method based on a large language model

By employing cross-modal knowledge alignment and adaptive prefix tuning strategies based on large language models, the problem of insufficient accuracy in cold load forecasting for green data centers under small sample conditions is solved, achieving high-precision and stable cold load forecasting, and supporting intelligent energy management and carbon emission reduction for green data centers.

CN122154752APending Publication Date: 2026-06-05ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ZHEJIANG UNIV
Filing Date
2026-01-22
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing data center cooling load forecasting methods lack accuracy in small sample sizes and under uncertain distribution environments, making it difficult to adapt to the dynamic changes of green data centers, resulting in significant energy waste and carbon emissions.

Method used

We adopt a large language model-based approach, which combines cross-modal knowledge alignment and adaptive prefix tuning strategies with a global interactive self-attention mechanism to achieve cross-modal feature consistency and multi-scale feature modeling, thereby improving the model's generalization ability and prediction accuracy under small sample conditions.

Benefits of technology

It improves the accuracy and stability of cooling load forecasting, reduces energy waste, and supports intelligent energy management and carbon reduction optimization for green data centers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122154752A_ABST
    Figure CN122154752A_ABST
Patent Text Reader

Abstract

The application discloses a green data center cold load prediction method based on a large language model. The method first constructs a multi-source operation data set, then constructs a data center domain knowledge base, and converts time series into text description, and fuses the two to obtain a context-aware template, realizing knowledge-enhanced semantic expression. Through a knowledge fusion alignment strategy, the time series features and text features are mapped to a unified latent space, ensuring cross-modal semantic consistency. An adaptive prefix tuning and global interactive self-attention mechanism is proposed to capture complex coupling relationships between devices and multi-scale dynamic features, thereby enhancing the model's generalization ability in dynamic environments. The method is significantly better than traditional deep learning models in terms of cold load prediction accuracy, stability and computational efficiency, and can effectively support intelligent energy management and carbon emission reduction optimization of green data centers, providing a new intelligent solution for low-carbon computing infrastructure.
Need to check novelty before this filing date? Find Prior Art