APPLICATION OF ARTIFICIAL NEURAL NETWORK FOR EVAPORATIVE COOLING IN DATA CENTRES

Abstract

A data center is a facility that may be used to house telecommunication or storage devices. Because of the 24/7 required operation of a data center, large segments of a data center are geared towards evacuating heat generated from operating one. Data centers entail multiple operating configurations, great amount of constraints and nonlinear correlations. The need to effectively optimize a data center operation presents a daunting challenge. The data center considered in this research is a test bed modular data center (MDC). Comprising of an Information Technology (IT), DEC and IEC module. Typical MDC are dynamic and complex in nature with various mechanical and electrical control systems aimed at continuous operation of the MDC. To achieve the aim of optimizing a data center, we propose the use of an Artificial Neural Network. A typical Artificial Neural Network architecture is dynamic in nature and able to perform adaptive learning in minimal computation time. An Artificial Neural Network model of the data center was created using its operating historical data. The Neural Network model allows for the ability to predict and control our MDC at optimum configuration. The MDC considered in this study is the MESTEX unit located in the Dallas Texas area. Using various parameters related to the operation of the unit. Such as Outside Air Temp, IT load, Cold Aisle Temp, Cold Aisle Humidity etc. We intend to give the Artificial Neural Network model some of these parameters as input and some as targets. In order to analysis and achieve real world results. The operational data used, are available from time periods logged on the MESTEK unit. More specific, the Neural Network model built in this study will be used to study weather impact analysis and prediction of future data (Step-ahead & Multi-Step ahead) that may be desired.