This paper presents a critical review of the state-of-the-art data-driven machine learning methods utilized for building energy forecast. Specifically, it offers a look into the advantages and disadvantages of four widely adopted machine learning methods: artificial neural networks, support vector machines, genetic algorithms, and decision trees. Based on the performance of these methods explored in previous studies, recommendations of application are provided for different categories such as building type (e.g., residential), forecasting method (e.g., long-term), and building energy (e.g., electricity). Some of the main identified research gaps include the lack of studies dedicated to long-term energy forecasts and inability to successfully incorporate occupant behavior into the models. This review also highlights the potential and prospects of hybrid models as avenues of growth in the domain of building energy forecast. Further research efforts in these areas of study can reap future benefits by promoting energy conservation thereby reducing the ecological footprint.
展开▼