Published
Privacy-preserving transfer learning framework for building energy forecasting with fully anonymized data was published in Applied Energy in June 2026.
Authors: Wonjun Choi (Chonnam National University), Sangwon Lee (Dartwork), Max Langtry and Ruchi Choudhary (University of Cambridge). Wonjun Choi and Sangwon Lee are co-first authors.
What it does
Transfer learning helps a building with little data borrow from other buildings — but choosing which buildings to borrow from usually needs metadata such as building type and size, which anonymized data does not have. The framework learns similarity directly from anonymized time series and uses it to pick sources.
Across 89 real-world buildings it reduced median forecasting error (MSE) by 27–31% and improved results in 99.2% of 356 transfer configurations, while needing only 0.51% of the communication bandwidth of federated learning.