Data Center Cooling System Energy Optimization
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)
Tool Stack
- Kafka, Python, Grafana
40
%Saving in Energy Consumption
60
%Increased Equipment Lifetime
25
%Increased Server Availability