Evolutionary double attention-based long short-term memory model for building energy prediction: Case study of a green building

Abstract

The prediction of building energy consumption plays a crucial role in building energy management and conservation because it contributes to effective building operation, energy efficiency evaluation, fault detection and diagnosis, and demand side management. Although a large number of energy prediction methods have been proposed, each method has its pros and cons and still exhibits the potential to be improved. This study proposes an evolutionary double attention-based long short- term memory model and introduces binary features by using feature combination. The proposed model is adopted to analyse the building energy consumption data of a green building in Shenzhen, China. The prediction performance of the proposed hybrid model measured via root-mean-square-error and mean absolute error are 4.02 and 2.87 respectively, which are evidently better than those of the base models. Results also show that an attention mechanism can improve the efficiency of the long short-term memory algorithm with which the model uses the input time series data. Meanwhile, binary features exert a significant effect on energy consumption. The proposed model is valuable to researchers and practitioners. It helps researchers apply artificial intelligence-based methods to building energy prediction from the perspective of paying selective attention to input data. Practitioners will benefit from developing accurate diagnosis of building energy efficiency and decision support for building retrofitting.