Predicting energy use in construction using Extreme Gradient Boosting.

Area of Science:

• Energy Engineering
• Computational Science
• Environmental Science

Background:

• Global energy consumption is rising due to economic and population growth.
• The construction industry accounts for a significant portion (20.1%) of global energy use.
• Accurate energy consumption estimation is vital for energy savings and mitigating global warming.

Purpose of the Study:

• To explore computational methods for estimating energy consumption in the construction sector.
• To evaluate the performance of machine learning-based approaches over traditional methods.
• To propose and validate the use of Extreme Gradient Boosting (XGB) for energy use prediction.

Main Methods:

• Utilized a dataset of hourly energy consumption from an office building in Shanghai (2015-2016).
• Implemented Extreme Gradient Boosting (XGB), a tree-based ensemble learning algorithm.
• Compared model performance using historical features only, date features only, and a combination of both.

Main Results:

• Machine learning-based frameworks demonstrated superior performance compared to statistics-based and engineering-based approaches.
• The XGB model incorporating both historical and date features outperformed models using single feature types.
• The best-performing XGB model achieved a Root Mean Square Error (RMSE) of 109.00 and a Mean Absolute Percentage Error (MAPE) of 0.24.

Conclusions:

• Extreme Gradient Boosting (XGB) is an effective method for predicting construction energy consumption.
• Combining historical and date features significantly enhances prediction accuracy.
• The developed model contributes to more efficient energy management and supports global warming reduction efforts.