University of Twente Student Theses

Login

Energy efficient control of data center HVAC systems using reinforcement learning

Six Dijkstra, Pepijn (2024) Energy efficient control of data center HVAC systems using reinforcement learning.

[img] PDF
13MB
Abstract:This thesis investigates the use of Reinforcement Learning (RL) to create a controller that can reduce the energy usage of the Heating, Ventilation and Air-Conditioning (HVAC) system in a Data Center (DC), while still meeting critical temperature constraints. The introduction highlights the importance of reducing energy in DC HVAC systems while exploring the current literature on the topic. Following the introduction comes the problem statement, where a simple HVAC system using a chiller plant is defined, of which a simulation model is created in EnergyPlus. This thesis will design its controller on this plant. Also, the goal of the controller of minimizing HVAC power while handling critical temperature constraints in a server room is formalized in an optimization problem. Then, the methodology chapter first describes the design of a baseline controller to compare with the RL-based controller. Finally, it proposes an RL framework with a tuneable reward function and consequently, algorithm and reward hyperparameter tuning experiments are set up. These experiments are used to tune the RL-based controller. All frameworks and experiments are implemented in Python. The numerical results chapter presents the results of the baseline and tuning experiments outlined in the method. From this, 3 tuned RL-based controllers are then selected and compared to the baseline. These 3 controllers are either good at handling constraints, reducing HVAC power and one with a combination of both. From this comparison, it follows that the RL-based controllers can outperform the baseline, but how these agents learn is still unpredictable, as they are sensitive to their initialization. In the conclusion, the key findings of this thesis are summarized, and the limitations of the simulation model and proposed controller are discussed. Next to that, directions for further research are proposed.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:52 mechanical engineering
Programme:Mechanical Engineering MSc (60439)
Link to this item:https://purl.utwente.nl/essays/103170
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page