Energy modeling and predictive control of environmental quality for building energy management using machine learning

Abstract

Heating, Ventilation, and Air Conditioning (HVAC) systems play a vital role in building energy management by controlling the indoor temperature and ensuring the occupant's comfort. However, the energy consumption of HVACs contributes significantly towards overall energy usage of a building and carbon footprint. To address this challenge, this research proposes the development of a predictive model for HVAC temperature forecasting using Machine Learning (ML) algorithms to optimize energy efficiency while maintaining thermal comfort in buildings. The study focuses on comparing the performance of Transformer Neural Networks and CNN-LSTM, a seq2seq model combining Convolutional Neural Networks (CNN) and Long-Short Term Memory (LSTM) on multiple forecasting horizons using data obtained from multiple devices deployed in a room verified by feedback survey forms filled by occupants. The transformer model outperformed, achieving an R2 score of 0.936 at a 1-minute forecasting horizon, surpassing the performance of CNN-LSTM model at all tested forecasting horizons. The transformer model yielded significant energy savings thereby reducing energy consumption by almost 50 % compared to the non-AI conventional methods, particularly at forecasting horizons of 1 min and 60 min, while the occupant survey also favoured a 60-minute forecasting horizon. The performance of transformer model particularly with a 60-minute forecasting horizon underscores its potential to optimize energy efficiency while ensuring thermal comfort in building energy management systems.