Nowadays, massive quantities of energy and, therefore money, are being wasted in data center refrigeration, as an effect of the lack of an effective data analysis system. In addition, a fact that must be considered, is that about a 2 % of the carbon dioxide emissions are caused by TIC services (in whose data centers are included).
The purpose of this project is to carry out a digital transformation based in data analysis and automatic prediction of variables, trying to reach a higher energy efficiency in data centers, aiming to save expenses in refrigeration, being this beneficial for the environment and the data centers owners. It is necessary to acquire knowledge about the current situation of the sector, and the main new technologies.
Artificial Neural Networks, built using Python, leaning on TensorFlow and Keras and libraries such as Pandas, are the main tools used to predict temperature reliably in every point of the data center, to adapt the refrigeration system along time, achieving that wage reduction. In this way, servers will be kept in the optimum temperature in order to assure a good efficiency, as they need to work in an adecuate range of temperature. The network is trained with measures taken by different sensors disposed in a center. Quality results are reached, obtaining a root mean squared error about 1,437 (this is the error measured in degrees).
Another mission is to study the most suitable number and disposition of those sensors. In this phase, similar results are obtained carrying out a reduction of the number of sensors and the measurements taken by each one.