Environment data processing for a data centre (1)

Abstract

Data centres play an essential role in today's global digital economy. Data centre construction has been surging worldwide due to digitalisation and digital solutions needs such as cloud technology. Data centres in Singapore consumed 7 per cent of the country’s total electricity in 2020. Hence, in Singapore, energy-efficient cooling methodologies such as air free cooling data centres are studied to determine their feasibility. An initial experimental result of an air-free cooled data centre testbed proves that air-free cooling may be feasible in Singapore, and machine learning models may be used to predict and improve the efficiency of the air-free cooling PID controller. Hence, researchers often seek methodologies that enhance these models. This project aims to study the dataset obtained from the testbed to determine if the accuracy of the machine learning model, namely the Neural Network, Decision Tree, Random Forest and Support Vector Machine model, may be improved by providing them with an additional feature generated using known physics law. The fan energy and airflow may be described using physics law, namely, fan law. The experiment implemented a polynomial regression model that references the fan law to predict the sum of fan energy, representing the additional feature input used by the other machine model. The experiment results show slight improvement for the Neural Network and Decision Tree Model. Hence, future work may focus on optimising the machine learning modal or the polynomial regression model to improve the accuracy further.