A real-time indoor temperature and relative humidity prediction method for exhibition hall based on Long Short-Term Memory network

Indoor temperature and relative humidity, along with traffic flows, were primary factors influencing the thermal environment within exhibition halls. Therefore, maintaining high-quality interior conditions during the operation stage was critical for both occupants and exhibits. This study introduced a real-time temperature and humidity prediction method for the operational stage of exhibition halls using a Long Short-Term Memory (LSTM) model. The LSTM model was trained on a dataset comprising 60 days of monitoring data from an exhibition hall in a museum located in Beijing, northern China. The Root Mean Square Error (RMSE), Mean Squared Error (MSE), and R2 values for indoor temperature, humidity, and traffic flow during the operational stage in the validation dataset were 0.0316, 0.0009, and 0.97, respectively. This trained model provided real-time predictions of temperature and humidity, assisting building managers in making informed environmental control decisions. The model was further validated on a new case involving public spaces with varying building areas and heights. The performance of the LSTM model was confirmed through the development of a temperature and humidity prediction tool. The output trends of the LSTM model were found to be consistent. Consequently, the proposed method effectively and swiftly reflected the variation in operational stage temperature and humidity in the next moment due to changes in the spatial geometry of the public hall, outperforming non-sequential prediction methods. © 2024 Elsevier Ltd