Hybrid model combining multivariate regression and machine learning for the rapid prediction of interior temperatures affected by thermal diodes and solar cavities. (1st March 2022)

• Record Type: Journal Article
• Authors: He, Yi; Kou, Fangcheng; Wang, Xin; Zhu, Ning; Song, Yehao; Chu, Yingnan; Shi, Shaohang; Liu, Mengjia; Chen, Xinxing
• Is Part Of: Building and environment. Volume 211(2022)
• Publication Date: 2022-03-01
• Subjects: Hybrid model -- Multivariate regression -- Machine learning -- Heat pipe thermal diode -- Solar cavity -- Rapid prediction; Buildings -- Environmental engineering -- Periodicals; Building -- Research -- Periodicals; Constructions -- Technique de l'environnement -- Périodiques; Electronic journals; 696
• DOI: 10.1016/j.buildenv.2021.108723
• Languages: English
• ISSNs: 0360-1323

Abstract

Abstract: Sustainable design often requires highly efficient building performance evaluations. This study proposed a hybrid model combining multivariate regression modelling (MRM) and machine learning modelling (MLM) for the rapid prediction of interior temperatures affected by heat pipe thermal diodes and solar cavities based on experimental data. A heat pipe thermal diode can promote unidirectional heat transmission from the solar cavity on the south side of our newly built experimental house to the indoor environment to increase the interior temperature and reduce the heating load in cold climates. Experimental data were collected and then imported, cleaned, and split according to MRM and MLM requirements, respectively. In MRM, linear multivariate formulas were generated according to the thermal diode's two different working conditions. In MLM, a machine-learning model was created and trained using the experimental data. The results our hybrid model produced were comprehensively evaluated via R-square, statistical discrepancies, and complex MRM analyses. The similarity between the prediction and experimental results clearly demonstrates our model's accuracy and efficiency. This research was an original attempt to integrate emerging computational tools and provide a means to perform highly efficient quantitative analysis of indoor thermal environments for environmental studies and sustainable designs in the early stages. Highlights: A hybrid model for rapid evaluation of indoor thermal environment was proposed. Multivariate regression and machine learning jointly formulated the hybrid model. Interior temperature affected by heat pipe thermal diode and solar cavity could be predicted in a very short time. Efficiency and accuracy of the model were demonstrated by benchmark investigations. Rapid evaluation on the built environment could benefit the development of sustainable architectural design.