Das Sharma, Shansita, Coursey, Austin, Quinones-Grueiro, Marcos, & Biswas, Gautam. (2024). Comparison of transfer learning techniques for building energy forecasting. In Proceedings of the 12th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes (SAFEPROCESS 2024), Ferrara, Italy, June 4-7, 2024, Volume 58, Issue 4, Pages 180-185. https://doi.org/10.1016/j.ifacol.2024.07.214
This study addresses the challenge of improving energy efficiency in buildings by accurately predicting normal and abnormal operations, which is crucial for effective energy management. A key obstacle in applying predictive models in real-world scenarios is the limited availability of data in many buildings, which makes it difficult to build reliable models. To tackle this, the authors investigate the use of three common transfer learning techniques, which allow models trained on one building to be adapted to others, even when there is less data available.
The study focuses on real-world data from internal building measurements and finds that transfer learning significantly enhances the accuracy of energy consumption models. By transferring knowledge between buildings, the models can better predict energy use and identify potential inefficiencies, even in buildings with sparse data. This approach offers a practical solution for improving energy management systems, especially in diverse building environments, where data limitations often hinder the effectiveness of traditional predictive models.
