Predicting the Extended Life Cycle Energy Consumption of Building Based on Deep Learning LSTM Model

Sustainable developments have been one of the main social forces worldwide, especially in the building sector. As the current biggest energy consumption industry of 35%, it is urgent to solve severe energy issues through advanced energy-saving technologies. Energy consumption prediction, as one of the important building energy management tools, can evaluate energy conservation policies and services timely. Unfortunately, there is still a significant difference between actual and predicted values. A consensus about the life cycle energy boundaries for buildings is being challenged. Some studies believed that the mobile energy related to building location can be accounted, except for traditional embodied and operational energies. Besides, deep learning was regarded as a method better than other simulation models in time series forecasts. To fill the gap between actual and predicted energy consumption values, this paper proposes to extend the life cycle energy boundaries of buildings and choose the Long Short-term memory model (LSTM)) to predict the building energy consumption in China from 2020 to 2029, which is based on the historical data collected from 2005 to 2019. Results show that there was a remarkable increase in the past 15 years for the total life cycle energy consumption of buildings, but afterwards it will fluctuate at around 1,050 Mtce because of potential influencing factors such as recyclable concrete and prefabricated process applied into an increasing number of newly built buildings. Mobile energy consumption accounted for 24% share of total energy consumption, but it is expected to fall significantly in the next decade. Overall, this study provides a pathway to help reduce building energy consumption prediction errors.