Modeling Indoor Thermal Environments in Detached Wooden Homes Using Linear Regression and Neural Networks
- 1The University of Tokyo, Faculty of Engineering, Department of Systems Innovation, Tokyo, Japan (jump.23.99@gmail.com)
- 2The University of Tokyo, Graduate School of Frontier Scineces, Department of Environment Systems, Kashiwa, Japan (ihara-t@k.u-tokyo.ac.jp)
- 3Sekisui House Innovation and Communication, Ltd., Tokyo, Japan (hashimoto097@sekisuihouse.co.jp)
- 4Tokyo Environmental Public Service Corporation, Tokyo Metropolitan Research Institute for Environmental Protection, Tokyo, Japan (tsunematsu.nobumitsu@gmail.com)
Heat-related fatalities, predominantly from heatstroke, commonly occur indoors within residential settings. Consequently, accurate prediction of indoor temperatures is crucial for preventing such fatalities. However, few studies have measured the indoor thermal environment of general existing residential buildings rather than newly built residential buildings, and few studies that have predicted the indoor thermal environment based on the measurement.
This study measured the indoor thermal environment of 17 detached wooden homes in two towns in Ota Ward, Tokyo Metropolis, Japan from July 25 to September 20, 2018, spanning a total of 58 days. Indoor measurements included room temperature and relative humidity, globe temperature, atmospheric pressure, window open/closed status, and air conditioner operating status. Outdoor measurements included outdoor temperature and relative humidity. In addition, the wind speed, precipitation, and cloud cover observed by the Japan Meteorological Agency were obtained. Based on these data, we developed models to predict indoor temperatures and wet-bulb globe temperature (WBGT). The prediction models were constructed using linear regression and a three-layer neural network. The input variables include measured or observed values and characteristics of the residential buildings, such as the age of the buildings, the number of floors, the floor on which the room is located, and the number of occupants.
The model predicted room temperature and room WBGT with mean absolute errors (MAE) of 1.0°C and 0.8°C for non-air-conditioned rooms. The accuracy of room temperature and WBGT prediction improved by 0.1°C when the model was trained using data for the predicted room collected during the previous period. The linear regression model was as accurate as the neural network. The factors that are important for predicting the indoor thermal environment were identified using linear regression.
This research enables the development of government-led alert systems to mitigate heat-related risks in residential areas.
How to cite: Li, Y., Hashimoto, Y., Tsunematsu, N., and Ihara, T.: Modeling Indoor Thermal Environments in Detached Wooden Homes Using Linear Regression and Neural Networks, EMS Annual Meeting 2024, Barcelona, Spain, 1–6 Sep 2024, EMS2024-685, https://doi.org/10.5194/ems2024-685, 2024.
