Smart Building Energy Management using Deep Learning Based Predictions

The prediction of electricity consumption in a building is critical for recognizing the possibilities for energy savings as a part of the digitalization of the built environment. This also helps to mitigate the effects of climate change, since buildings are required to be more adaptable and resilient while consuming less energy and maintaining user comfort. Peak energy demand may be detected using historical building data, allowing users to more efficiently manage their energy consumption while also providing the demand side management response to the utilities for the necessary control and actuation in real-time. In view of this, the paper focuses on various deep learning methods (re-current neural network (RNN), long short term memory (LSTM), and gated recurrent unit (GRU)) to predict electricity consumption of three different types of buildings in a model-free environment. A hybrid model is also developed by combining the features of RNN and GRU for predicting the load profile. Another major contribution of the paper is the introduction of hyperparameter tuning for improving prediction accuracy. The results highlight the effectiveness of the hybrid model in predicting electricity consumption and also show the improvement in prediction accuracy using hyperparameter tuning.