Optimizing the Operation of Grid-Interactive Efficient Buildings (GEBs) Using Machine Learning

The building sector constitutes 40% of global electric energy consumption, making it vital to address for achieving the global net-zero emissions goal by 2050. This study focuses on enhancing electric load forecasting systems’ performance and interactivity by investigating the impact of weather and building usage parameters. Hourly electricity meter readings from a Texas university campus building (2012–2015) were employed, applying pre-processing techniques and machine learning algorithms such as linear regression, decision trees, and support vector machines using MATLAB R2023a. Exponential Gaussian Process Regression (GPR) showed the best performance at a one-year training data size, yielding an average normalized root mean square error (nRMSE) value of 0.52%, equivalent to a 0.3% reduction compared to leading methods. The developed system is presented through an interactive GUI and allows for prediction of external factors like PV and EV integration. Through a case study implementation, the combined system achieves 12.8% energy savings over a typical year simulated using ETAP 22 and Trimble ProDesign software version 2021.0.19. This holistic solution precisely models the electric demand management scenario of grid-interactive efficient buildings (GEBs), simultaneously enhancing reliability and flexibility to accommodate diverse applications.