Predictive Cooling Control for Data Center Power Swings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cooling systems for data centers are inefficient due to their reliance on adjusting cooling based on current operating conditions, leading to unnecessary resource overuse when conditions change rapidly, resulting in suboptimal power consumption management.
Innovation Solution
A method using a machine learning model to estimate forthcoming power consumption of data processing devices, combined with a cooling-control model that adjusts cooling system operations based on current power consumption, weather conditions, and safe operating temperatures, allowing for more efficient control of the cooling system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cooling system operation is adjusted based on current operating conditions, then the cooling system responds to immediate thermal demands, but unnecessary resource overuse occurs when operating conditions change rapidly
Solution Approach 1:
The system performs preliminary actions by using a machine learning model to predict future power consumption of data processing devices before it actually occurs. This prediction enables the cooling control system to anticipate thermal demands and adjust cooling operations in advance, avoiding the need to react to rapid changes in operating conditions and thereby reducing unnecessary energy consumption.
Solution Approach 2:
The system dynamically adapts cooling control by continuously updating predictions based on changing operating conditions. The machine learning model processes real-time data about data processing device operations, environmental conditions, and historical patterns to generate dynamic predictions that allow the cooling system to optimize its operation continuously rather than relying on static or reactive control.
2Temperature
If cooling is adjusted based on current power consumption, then immediate cooling needs are met, but power usage effectiveness deteriorates due to inability to anticipate future power needs
Solution Approach 1:
The system performs preliminary cooling adjustments by predicting future power consumption before it occurs. The machine learning model analyzes patterns in data processing device operation, environmental conditions, and historical data to forecast upcoming power demands, enabling the cooling system to proactively adjust and maintain optimal temperatures while minimizing energy consumption.
Solution Approach 2:
The system implements a feedback mechanism where actual power consumption and temperature data are continuously monitored and fed back to the machine learning model. This feedback loop allows the model to refine its predictions and the cooling control system to continuously optimize its operations based on the difference between predicted and actual conditions, improving overall power usage effectiveness.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for controlling a cooling system operatively connected to a data processing device for cooling thereof. The method is executed by a controller operatively connected to the cooling system. The method comprises determining, using a machine learning model executed by the controller, an estimated forthcoming power consumption of the data processing device, the machine learning model being based at least in part on a history of power consumption of the data processing device, determining, using a cooling-control model executed by the controller, at least one control signal for controlling the cooling system, determining the at least one control signal being based at least in part on the estimated forthcoming power consumption from the machine learning model, and controlling, by the controller, the cooling system based on the at least one control signal for controlling the cooling system.