QUANTIFYING UNCERTAINTY OF REAL-TIME ENERGY MODEL AND ITS APPLICATION

Abstract

Buildings are responsible for a large percentage of the world’s energy consumption and this is expected to rise. Recognising the huge impact that electricity production and consumption has on the environment, causes energy management and efficiency to become increasingly important. With the availability of data from smart meters, big data analytics and machine learning are widely used to make sense of these data such that better decisions can be made to improve energy efficiency.

A real-time energy model was developed using the Light Gradient Boosting Machine algorithm to predict electricity consumption in this study. The Building Data Genome Project II dataset, which is available online, provides two years of data in hourly resolution. The data from the target building was split into one year for training data (2016) and one year for test data (2017).

Feature engineering occurred in three stages and performance evaluation was done using the statistical metrics of R2 value, mean absolute error (MAE) and mean absolute percentage error (MAPE) after each stage. The resultant model has a high R2 value of 0.945 alongside low MAE of 12.18 and low MAPE of 3.11%, signifying an excellent fit of model on the target building.

As a high goodness of fit was observed in the target building, the model was applied to other buildings belonging to the same site of the target building to test its robustness. R2 values were obtained for all buildings on the site for quantiles and statistics analysis. After data cleaning, the 95% confidence interval lies from 0.682 to 0.787, suggesting a decent fit of the model.

Quantifying uncertainty of the real-time energy model helps the energy management team make better decisions to enhance energy efficiency within the buildings. The application of fault detection and diagnosis (FDD) using the real-time energy model was explored. The hybrid FDD can leverage on the highly accurate predictions to identify anomalies as faults and diagnose them according to the simulated faults database. This would result in cost savings due to efficient energy management of buildings. Furthermore, enhanced energy management minimises the impact on climate change, taking a step closer to achieve sustainability.