Machine Learning-Based Carbon Compliance Forecasting and Energy Performance Assessment in Commercial Buildings

Owing to the need for continuous improvement in building energy performance standards (BEPSs), facilities must adhere to benchmark performances in their quest to achieve net-zero performance. This research explores machine learning models that leverage historical energy data from a cluster of buildings, along with relevant ambient weather data and building characteristics, with the objective of predicting the buildings’ energy performance through the year 2040. Using the forecasted emission results, the portfolio of buildings is analyzed for the incurred carbon non-compliance fees based on their on-site fossil fuel CO 2e emissions to assess and pinpoint facilities with poor energy performance that need to be prioritized for decarbonization. The forecasts from the machine learning algorithms predicted that the portfolio of buildings would incur an annual average penalty of $31.7 million ($1.09/sq. ft.) and ~$348.7 million ($12.03/sq. ft.) over 11 years. To comply with these regulations, the building portfolio would need to reduce on-site fossil fuel CO 2e emissions by an average of 58,246 metric tons (22.10 kg/sq. ft.) annually, totaling 640,708 metric tons (22.10 kg/sq. ft.) over a period of 11 years. This study demonstrates the potential for robust machine learning models to generate accurate forecasts to evaluate carbon compliance and guide prompt action in decarbonizing the built environment.