Determining the heat transfer coefficient of sustainable buildings using extreme learning machine

Abstract

In an attempt to eliminate the performance gap in buildings, recent studies have demonstrated the value of on-site heat loss measurements. However, it is recognized that its measurement setup and calculating processes are difficult and complicated. Heat loss via party walls to the outdoors (as opposed to the neighboring dwelling) has drawn a lot of attention, and the building regulations failed to adequately address this issue. In the jamb sections where window frames meet the wall, there is a significant loss of heat. The heat conduction matrix analysis was done using an extreme learning machine (ELM). The main objective for the heat transfer design may be created by a quick and accurate prediction of the building's energy usage during the early design stage. Due to its speed, accuracy, and ability to handle nonlinear variables, ELM is the most popular artificial intelligence model for optimizing building performance. In this study, an approach for predicting building heat loss was developed to simplify building shapes early in the design process and figure out the heat loss ratio. Regarding the regression indices analysis (R2 = 0.879, RMSE = 0.633, r = 0.743), ELM could prove itself as a reliable model in this analysis. In terms of average absolute variation values, ELM demonstrated the highest level of accuracy (85.08 %). Furthermore, ELM required 0.125 s of CPU time to achieve a testing root mean square error (RMSE) of 0:0094, whereas the BP algorithm required 21.21 s of CPU time to achieve a much higher testing error of 0.0151. When compared to the traditional BP models, the new ELM operates 175 times faster. For the wall-foundation system and the slab-on-ground structure, the findings of the heat flux and heat flow acquired using dependable models are 4.3 % and 6.1 % and 10.3 % and 10.7 %, respectively, higher than those obtained using base models. As a consequence, there was a significant difference between the entire house heat loss coefficients that were projected and observed. It was revealed that the bypassing of the thermal insulation through the uninsulated party wall spaces was the most probable reason for the difference.