Date

2022-7

Author

YILMAZ, DERYA

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

404
views

229
downloads

Cite This

The energy performance gap presents obstacles for governments to reduce dependency on foreign energy sources, as well as for policymakers to be successful in future ambitions about reducing the environmental impacts of buildings. Although considerable research focused on the reasons, reduction, or quantification of the energy performance gap in buildings, little attention has been given in the past to methods that would inform decision makers about a likely performance gap. Such studies could change the actions and strategies of decision makers to achieve expected energy performance levels in buildings. This research supports the idea that performance gap occurs due to project risks occurring during the building life cycle. Therefore, the research examined the energy performance gap in buildings through the lens of project risks and demonstrated the potential of the machine learning classification technique in the prediction of performance gaps. To achieve the aim of the research, first, a conceptual performance gap framework was proposed after a literature review and conducting semi-structured interviews with 11 experts working on six energy-efficient buildings. Later, the framework was designed as a web-based survey for data collection. Information about the performance gap and project risks of a total of 77 buildings was collected using the web-based survey. Lastly, using this dataset, the performance of the four machine learning classification algorithms (1) Naïve Bayes, (2) KNN, (3) SVM, and (4) Random Forest were compared to find the best prediction model based on six performance metrics (Accuracy, kappa statistic, precision, recall, F-measure, and AUC). Energy performance gap prediction problems were studied focusing on the heating demand and electricity demand of buildings. First, datasets were studied to distinguish cases that might show a positive and negative performance gap (binary classification problem). Following that, smaller datasets were studied to predict the level of performance gap in buildings using the following categories, low (0-15%), medium (15.1-40%), and high (40.1-90%), (multi-class classification problem). The results indicate that Naïve Bayes gives the best classification accuracy on the majority of the classification problems. While KNN shows the highest accuracy for electricity performance gap prediction (79.00%), Naïve Bayes is the suggested algorithm for heating performance gap prediction (72.50% accuracy) to solve binary classification. Moreover, in multi-class classification, while Naïve Bayes achieved an accuracy of 77.08% for positive electricity performance gap, SVM achieved an accuracy of 76.04% for positive heating performance gap prediction. Lastly, in multi-class classification, Naïve Bayes provided an accuracy of 83.85% and 71.81% to solve the negative electricity performance gap and heating performance gap, respectively.

Subject Keywords

Energy Performance Gap, Machine Learning, Classification, Algorithms, Risk Identification

URI

https://hdl.handle.net/11511/98639

Collections

Graduate School of Natural and Applied Sciences, Thesis