A hybrid building thermal modeling approach for predicting temperatures in typical, detached, two-story houses. (15th February 2019)

• Record Type: Journal Article
• Authors: Cui, Borui; Fan, Cheng; Munk, Jeffrey; Mao, Ning; Xiao, Fu; Dong, Jin; Kuruganti, Teja
• Is Part Of: Applied energy. Volume 236(2019)
• Page Start: 101
• Page End: 116
• Publication Date: 2019-02-15
• Subjects: Building demand management -- Data-driven model -- Supervised machine learning -- Particle swarm optimization; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2018.11.077
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: An indoor temperature prediction solution for multiple-zone houses is provided. Black-box methods can efficiently predict the dynamic building thermal performance. The prediction of average temperatures in downstairs and upstairs is accurate. Proposed modeling approach can be easily applied in different scenarios and houses. Abstract: Within the residential building sector, the air-conditioning (AC) load is the main target for peak load shifting and reduction since it is the largest contributor to peak demand. By leveraging its power flexibility, residential AC is a good candidate to provide building demand response and peak load shifting. For realization of accurate and reliable control of AC loads, a building thermal model, which characterizes the properties of a building's envelope and its thermal mass, is an essential component for accurate indoor temperature or cooling/heating demand prediction. Building thermal models include two types: "Forward" and "Data-Driven". Due to time-saving and cost-effective characteristics, different data-driven models have been developed in a number of research studies. However, few developed models can predict temperatures in respective zones of a multiple-zone building with an open air path between zones e.g., an open stairwell connecting two floors of a home. In this research, a novel hybrid modeling approach is proposed to predict the average indoor air temperatures of both the upstairs and downstairs. This "hybrid" solution integrates both gray-box, i.e. RC model and black-box models. A developed RC model is used to predict the building mean temperature, and black-box model, in which the supervised machine learning algorithms are leveraged, is used to predict the temperature difference between the downstairs and upstairs. Compared with the measured data from a real house, the results obtained have acceptable/satisfactory accuracy. The method proposed in this study integrates the advantages of black-box and gray-box modeling. It can be used as a reliable alternative to predict the average temperatures in respective floors of typical detached two-story houses.