Soft-sensor model for indoor temperature prediction under heating conditions

Abstract

Temperature information is important in everyday life. However, the local temperature at a certain point cannot be determined directly without using a physical thermometer, which can often disturb activity at that point and consume additional energy. Moreover, it would be costly to equip a room with many thermometers simply to measure local temperature. To overcome this problem, this study developed a model for prediction of local indoor temperature under heating conditions based on the heat transfer process mechanism using the soft-sensing method. To simulate the heating of air conditioning in winter, water at temperature of 30/50 °C was fed into water at 10 °C in an acrylic box. The soft-sensor model was trained and validated with transient temperature data and the multiple linear regression method. The particle image velocimetry method was used to obtain the velocity field. Temperature inhomogeneity in the vertical direction was evident for the feed temperature of 50 °C. The temperature variation in the bottom layer was divided into the dead time, rising, and settling periods, which was caused by variation of flow state. Multiple evaluation indicators suggested that the model performed with acceptable accuracy for different feed temperatures and locations. The bottom layer model at 50 °C required correction for the dead time, and model accuracy was improved by separately modeling the rising and settling periods. This soft sensor could be installed in air-conditioning systems to achieve optimal control, provide user comfort, and save energy.