Accurate Building Energy Consumption Prediction with Convolution Recurrent Deep Neural Networks

Abstract

Accurate energy consumption prediction is a prerequisite for effectively dispatching distributed power sources. For a building, due to the frequent fluctuations derived from many dynamic factors, the precise energy consumption prediction is still facing challenges. Existing methods usually only use common recurrent neural networks to predict building energy consumption, consider common recurrent neural networks model does not have the ability to extract spatial features and they have a long-term memory problem, so they have limitations to deal with long term task. To overcome these challenges, in this paper, we propose a hybrid model to predict the cooling consumption of a building.Our hybrid model has the merits of convolutional neural network and gated recurrent unit in capturing spatial-temporal features. Experiment results show that our hybrid model has the best performance, compared with other methods. The result will benefits managers to make reasonable scheduling of power and equipments.