Predicting Energy Consumption of Office Buildings: A Hybrid Machine Learning-Based Approach

Abstract

Improving building energy efficiency requires an understanding of the affecting factors and an assessment of different design and operation alternatives. In this context, accurate prediction of building energy consumption gained a lot of research attention. In recent years, a significant number of building energy consumption prediction models, with various intended uses, have been proposed. However, existing data-driven models are mostly based on outdoor weather conditions, but do not take occupant behavior into account. Towards addressing this research gap, this paper presents a hybrid machine-learning and data-mining approach to develop prediction models that learn from both real data and simulation-generated data. Real data were collected from an office building, including data about building energy consumption, outdoor weather conditions, and occupant behavior. Simulation-generated data were created through simulating an office building in EnergyPlus. A feature selection algorithm was used to determine the critical features in predicting energy consumption for office buildings. A set of regression models were then trained for predicting the hourly values of an outdoor weather-related factor and an occupant behavior-related factor based on these features. Then, an ensembler model—which takes the outputs of the outdoor weather-related factor and occupant behavior-related factor models—was trained to predict cooling energy consumption. In training the models, several machine learning algorithms—such as Gaussian Process Regression (GPR), Support Vector Regression (SVR), Artificial Neural Networks (ANN), and Linear Regression (LR)—were tested. The predicted energy consumption levels showed agreement with the actual levels. This indicates that the proposed regression models can help support decision making related to office buildings.