Data Center Cooling Load Prediction and Analysis based on Weather Data Clustering

Abstract

With the rapid development of data center, it is urgent to reduce energy consumption from the perspective of central cooling plant. At the same time, with the deepening application of clustering methods in data analysis, this study combines the K-means clustering method in machine learning with the energy consumption simulation software, DeST and applies it to actual case. The weather data of whole year are clustered into 29 typical daily patterns in Changzhou to study the load characteristic. It is found that there are 9 operation modes in the data center. The impact of hourly weather data changes on the external load of the data center is analyzed. The positive and negative impact of the temperature on the load of the following day are 4.96 % and -5.73 % in the heating season, 3.72 % and -2.63 % in the cooling season, which can be ignored. The cooling load of IT equipment accounts for a large proportion while the external load of hot and cold only accounts for 5.02 % and -5.73 % in the data center. Due to its 24-hour operation, the annual load change is relatively stable. The accuracy of load prediction is 56.60 %.