Short-term forecasting model for residential indoor temperature in DHS based on sequence generative adversarial network

Abstract

With the rapid development of the economy and the continuous improvement of people's living conditions, building thermal comfort has become one of the essential objectives of the development of the smart district heating system (SDHS). The accurate prediction approach of indoor temperature is the primary prerequisite and basis for achieving optimal thermal comfort regulation. However, the buildings' indoor temperature has significant thermal inertia and nonlinear characteristics due to the influence of multiple factors. The traditional time-series prediction algorithm can hardly accurately extract the indoor temperature variation pattern and cannot fully meet the satisfactory regulation requirements of SDHS. Therefore, an indoor temperature prediction model based on a sequence generative adversarial network (SGAN) is proposed in this paper. The new SGAN algorithm is trained by iterative adversarial training of the generator and discriminator, and the LSTM model built into the generator can effectively extract the high-level nonlinear abstract features of indoor temperature to achieve its accurate prediction. The detailed comparative experimental results show that the proposed indoor forecasting algorithm based on SGAN has obvious performance advantages compared to state-of-the-art algorithms, such as random forest regression (RFR), gradient boosting regression (GBR), support vector regression (SVR), adaptive boost (AdaBoost), multilayer perception (MLP), and long-short term memory(LSTM). The SGAN's mean absolute percentage error (MAPE) index reaches 2.3%.