Optimization of Server Room HVAC Systems for Energy Efficiency. Leveraging CFD and AI-Driven Techniques
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Engineering mathematics and computational science (MPENM), MSc
Publicerad
2024
Författare
Carlsson, John
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
The increasing demand for energy efficiency in server rooms and Heating, Ventilation & Air Conditioning (HVAC) systems necessitates advanced optimization techniques to reduce energy consumption while maintaining thermal stability. This thesis focuses on holistic energy optimization of a server room’s HVAC system by optimizing the room’s geometry and HVAC parameters using computational methods such as CFD and machine learning models, including an Artificial Neural Network (ANN). The study employs a range of meta heuristic optimization algorithms, including Genetic Algorithm (GA), Simulated Annealing (SA), Particle Swarm Optimization (PSO), and Differential Evolution (DE), to identify the best configuration for reducing
energy usage.
The goal of the optimization process is to maximize the inflow air temperature and minimize the mass flow while ensuring that the average room temperature does not exceed 28°C. The project was conducted at a worst case scenario, when the outside air temperature is 30 degrees Celsius and the server room is running on maximum
capacity, generating waste heat of 72kW.
A Latin Hypercube Sampling (LHS) method was employed to capture the underlying differential equations behavior throughout the high-dimensional parameter space, and a neural network was trained on this sample data to predict room temperature. Particle Swarm Optimization was used in order to find the optimal parameters for
minimal energy consumption.
The results demonstrate that the proposed optimization techniques can significantly
enhance the energy efficiency of HVAC systems in server rooms at a worse case scenario.
By adjusting the inflow temperature and air mass flow, a reduction in cooling
energy consumption of up to 22.69% was achieved. Future work could include hybrid
optimization approaches to further improve system performance and the application
of multi-objective optimization to accommodate varying operational phases.
Beskrivning
Ämne/nyckelord
CFD , Optimization , Machine learning , Neural network , energy efficiency