Building energy consumption prediction plays an irreplaceable role in energy planning and conservation as it is an important part of a building's lifecycle management process, since it allows organizations to make decisions about what investments need to be made in order to reduce energy consumption. Data-driven approaches, such as artificial neural networks, support vector regression, support vector machine, KNN, gradient boosting regression, and extreme learning machine are the most advanced methods for building energy prediction. However, owing to the high nonlinearity between inputs and outputs of building energy consumption prediction models, these approaches require improvement regarding the accuracy of predictions, robustness (i.e., the ability to produce accurate results in situations where there is a high degree of uncertainty or variability), and generalization ability (i.e., the ability to accurately predict outcomes when presented with new data). To counter these shortcomings, an ensemble learning method is proposed, which will use both SVR and SVM. It combines these individual models together to improve the stability and predictive power of the model through learning several simple models and combining their output to produce the final decision.