Building heating management based on deep learning to strength intelligent building design

Building heating systems based on deep networks plays an important role in intelligent urban construction, providing a comfortable environment for residents. An accurate heating prediction system which collects time-series signals through various sensors can grasp the temperature of buildings in advance, avoiding unnecessary energy waste. Traditional convolution operations only model the local dependency relationships of time-series signals for temperature prediction. To improve the long-range dependencies of features, long short-term memory networks (LSTM) and Transformer-based models have been widely used in building heating prediction tasks. However, local dependencies cannot be utilised, and these models require a large amount of computation. To this end, this paper proposes a novel variant Swin-Transformer-based model by exploiting a sparse self-attention mechanism for building temperature prediction. In addition, we introduce the traditional convolution operations for local dependency modelling. We conducted extensive experiments on our self-built dataset. All experimental results show that the proposed model achieves higher performance than LSTM and Transformer models in terms of temperature prediction.