Volume 13: Proceedings Applied Energy Symposium: CUE2020, Part 2, Japan/Virtual, 2020

A transfer learning method for building energy prediction using long short term memory and domain adversarial neural network Xi Fang, Guangcai Gong, Liang Chun, Wenqiang Li, Pei Peng


To overcome the data shortage problem of model training, this study proposes a novel transfer learning strategy for short term cross-building energy prediction using long short term memory (LSTM) and domain adversarial neural network (DANN). The proposed strategy can utilize transferred knowledge learnt from related domains with sufficient historical data. LSTM based feature extractor is used to extract temporal features across source and target domains. DANN attempts to find domain invariant features between the source and target domains via domain adaptation. Then, the domain adaptation based transfer learning model (i.e. LSTM-DANN) trained with data from different buildings can be directly applied to predict the target building energy without having its prediction performance degradation caused by domain shift. Experiments are conducted to evaluate the performance of the proposed transfer learning strategy in different scenarios. Results demonstrate that domain adaptation can well overcome the domain shift between the source and target domains by learning the domain invariant features. Furthermore, the proposed strategy can significantly enhance the building energy prediction performance compared to models trained on the target only data, the source only data, both the target and source data, but without domain adaptation.

Keywords transfer learning, long short term memory, domain adversarial neural network, building energy prediction

Copyright ©
Energy Proceedings