The building sector is responsible for excessive amounts of energy consumption and energy related emissions. The need to implement sustainable technology is crucial, such as heat pumps. However, faults within the system puts a strain on energy consumption. Studies have revealed that digital monitoring has the possibility to enhance energy-efficiency. Specifically, Machine Learning techniques has proven to be successful in the field of system optimization and fault detection which can help reduce energy consumption and additional costs. The aim of this thesis is to investigate how data analysis and Machine Learning techniques can be used to make performance predictions of heat pump systems.This thesis focuses on developing regression models for heat pump performance predictions using data from real-time field measurements from two ground-source heat pumps connected to a large facility. The predicted performance metrics are the Coefficient of Performance (COP) and Carnot efficiency. The research methodology includes a literature review, data pre-processing, feature selection and model development. The data was pre-processed to remove unsatisfactory information, and relevant input features were identified during feature selection. Regression models were developed ranging from simple linear to non-linear regression with up to four input features and third-degree polynomial. The models were evaluated using three different model evaluation metrics.The results of this thesis revealed that the best performing models for predicting COP were non-linear, including three to four input features with a third-degree polynomial. These models were able to achieve over 90% accuracy. However, models predicting Carnot efficiency showed deviations for one of the heat pumps. These results revealed the importance of the feature selection process when developing regression models. The significant features to consider were revealed to be the discharge refrigerant (the temperature of the refrigerant at compressor outlet) and the compressor voltage signal. The data analysis process and regression models revealed assumed measurement error or faults in one of the heat pump systems.In conclusion, this study emphasizes the effectiveness of Machine Learning techniques for heat pump performance predictions, specifically highlighting the role of feature selection.
Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:kth-347863 |
Date | January 2024 |
Creators | Carlsson, Filippa |
Publisher | KTH, HÃ¥llbara byggnader |
Source Sets | DiVA Archive at Upsalla University |
Language | English |
Detected Language | English |
Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
Format | application/pdf |
Rights | info:eu-repo/semantics/openAccess |
Relation | TRITA-ABE-MBT ; 24384 |
Page generated in 0.0018 seconds