Building's Energy Consumption Prediction with Limited Historical Data via AR-RNN

Yilong Guo, Qiyu Wei, Yaxin Zhu, Qiang Wang, Yanan Liu

Abstract: As urbanization continues to spur the construction of buildings, managing energy consumption has become a crucial component of sustainable development. To increase building energy efficiency and achieve carbon neutrality, accurate and efficient energy consumption estimates are essential. This involves predicting future energy consumption based on historical data, developing energy efficiency control strategies, and reducing energy consumption and costs. However, newly constructed commercial buildings often have limited historical data, making it challenging to predict future power consumption accurately. To this end, we propose a series of data preprocessing methods, including dimension reduction and data interpolation, designing an Autoregressive Recurrent Neural Network (AR-RNN) based model with collected data from various equipment in the building. In the experimental results, our approach accurately anticipates energy usage for the following week, achieving a Mean Absolute Percentage Error (MAPE) score of 5.72%. The datasets and codes are available from: https://github.com/GuoYL125/Building-Energy-Consumption-Prediction-with-Limited-Historical-Data.git.

External IDs:dblp:conf/icit2/GuoWZWL24