Cooling Power Consumption Dependency Simulation for Real-World Data Center Data

Abstract

Data centers (DC) are considered one of the top electricity consumers and there is an increasing interest in finding more sustainable approaches to their power supply needs. Purchase of renewable energy is the straightforward approach but involves large investments and long-term commitments with a need for DC power consumption planning. In this paper, we propose a data-driven DC cooling power consumption prediction method that uses IT equipment power consumption and outdoor air temperature data as input features. The contributions of this work are the design of a differenced data-based modeling method with baseline value calculation and performance testing based on real-world data from air-based and liquid-based cooling DCs. In the experiments, we compare our proposed method with a raw data-based approach and identified the proposed method as the more stable performing one. The results show that the proposed method performs better when training data is limited and can handle sudden value drifts in the consumed cooling power caused by changes in cooling devices' operation settings with baseline adaptation.