Our client is KocSistem (part of Koc Group, a Fortune Global 500 company)
Problem
Data centers generate significant amount of heat that must be removed to keep the servers running. Cooling is a high energy consuming operation
Suboptimal temperature setpoints (overcooling and undercooling) creates inefficiencies
Data centers have complex, often non-linear thermal dynamics which complicates the cooling process
Action
The approach combined prediction and optimization models
Neural network models were implemented to predict the aftermath of temperature set point changes. Optimization models were used to determine the optimal set point
This dynamic decision making allowed the environment to adapt changes in the computational load and external factors (i.e. weather conditions)