Works
Papers and projects. Where our methods meet real problems.
Privacy-preserving transfer learning framework for building energy forecasting with fully anonymized data A novel framework that enables effective transfer learning using exclusively anonymized time-series data, achieving median MSE reductions of 27–31% across 89 real-world buildings while requiring only 0.51% of federated learning's communication bandwidth. Representation vector-based time series similarity analysis using unsupervised contrastive learning time series encoder Measuring how alike buildings' energy use is with TS2Vec representation vectors — a step toward choosing transfer-learning sources automatically. Dynamic management of ground thermal response uncertainty through temporal analysis of parameter sensitivity Managing the uncertainty of ground thermal response over time by analyzing how sensitive the response is to each parameter as the test progresses. Interpretable deep learning model for load and temperature forecasting: Depending on encoding length, models may be cheating on wrong answers An interpretable forecasting model for building load and temperature — and the finding that, depending on encoding length, a model can reach good scores for the wrong reasons. Real-time probabilistic backfill thermal property estimation method enabling estimation convergence judgment Real-time probabilistic estimation of backfill thermal properties for underground transmission lines and ground heat exchangers, with a convergence test that tells the field crew when the experiment can stop. Performance evaluation of deep learning architectures for load and temperature forecasting under dataset size constraints and seasonality How well do deep learning architectures forecast building load and temperature when data is scarce and seasons change? A side-by-side evaluation under dataset-size limits and seasonality.