Data center energy consumption perception quantization method and system facing algorithm and electricity collaboration

Abstract

This invention discloses a data center energy consumption perception and quantification method and system for computing-power collaboration, belonging to the field of energy consumption calculation technology. The method includes: classifying deep learning task types in the data center environment; modeling the energy consumption of each classified deep learning task to obtain a deep learning task energy consumption model, using sparse regression for prediction, obtaining a deep learning task energy consumption model based on sparse regression prediction; constructing a data center-level energy consumption quantification model based on the obtained deep learning task energy consumption model based on sparse regression prediction and the energy consumption of the infrastructure, and then formulating a computing power allocation strategy according to the data center's energy budget. This invention, through accurate identification and classification of deep learning tasks, can adopt specialized energy consumption modeling methods for different types and architectures of tasks, improving the accuracy and relevance of energy consumption assessment.

Images