A SHORT-TERM INDOOR TEMPERATURE PREDICTION MODEL BASED ON GCN-BILSTM-MHA

Accurate indoor temperature prediction can not only improve user comfort, but also effectively reduce energy consumption. In order to improve the accuracy of indoor temperature prediction, this study proposes a short- term indoor temperature prediction model based on GCN- BILSTM- MHA. The graph convolutional neural network (GCN) and bidirectional long short- term memory neural networks (BILSTM) were used to obtain the spatial- temporal features of the data, and the multi- head attention mechanism (MHA) was integrated to capture the important information and features of different positions and angles. The experimental results show that the four evaluation indexes of GCN- BILSTM- MHA model are superior to the comparison model. R^2 reached 99.2%. The model effectively improves the prediction accuracy of indoor temperature and provides some reference value for further improving the timeliness of intelligent heating.