Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Short-term forecasting of building energy consumption with deep generative learning

Yichuan X. Ma
The University of Hong Kong, Hong Kong S.A.R. (China)

Abstract: Short-term forecasting of building energy consumption is highly valuable from both technical and economic point of views. In this paper, a deep generative learning method taking account of short-term future meteorological data is proposed to forecast building energy consumption in the next 24 hours. A conventional multilayer perceptron and a non-meteorology version of the proposed GAN-based model were developed and comparatively tested as baseline models. Multi-year hourly meteorological data and actual energy consumption measurements from two office buildings in Shanghai were used for modelling and testing. The proposed model significantly outperformed the baseline models in all granularity settings. Decent cross-case generalisability of the proposed GAN-based models were demonstrated.
Keywords: Deep learning, short-term forecasting, building energy consumption, generative adversarial network
Pages: 2248 - 2253