A Method for Selecting the Appropriate Source Domain Buildings for Building Energy Prediction in Transfer Learning: Using the Euclidean Distance and Pearson Coefficient
Building energy prediction faces challenges such as data scarcity while Transfer Learning (TL) demonstrates significant potential by leveraging source building energy data to enhance target building energy prediction. However, the accuracy of TL heavily relies on selecting appropriate source buildin...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-07-01
|
| Series: | Energies |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1996-1073/18/14/3706 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Building energy prediction faces challenges such as data scarcity while Transfer Learning (TL) demonstrates significant potential by leveraging source building energy data to enhance target building energy prediction. However, the accuracy of TL heavily relies on selecting appropriate source buildings as the source data. This study proposes a novel, easy-to-understand, statistics-based visualization method that combines the Euclidean distance and Pearson correlation coefficient for selecting source buildings in TL for target building electricity prediction. Long Short-Term Memory, the Gated Recurrent Unit, and the Convolutional Neural Network were applied to verify the appropriateness of the source domain buildings. The results showed the source building, selected via the method proposed by this research, could reduce 65% of computational costs, while the RMSE was approximately 6.5 kWh, and the R<sup>2</sup> was around 0.92. The method proposed in this study is well suited for scenes requiring rapid response times and exhibiting low tolerance for prediction errors. |
|---|---|
| ISSN: | 1996-1073 |