Date

2021-06-29

Author

Duran, Ayça
Gürsel Dino, İpek

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

167
views

72
downloads

Cite This

Educational facilities account for approximately 12% of the energy consumed by buildings in the US and UK. Classrooms should provide their occupants' satisfactory indoor environments as indoor conditions play a determinant role in the performance, productivity, attendance, and health of students and teachers. Indoor air quality and thermal comfort are two major determinants of healthy classrooms. Generally, classrooms operate at full capacity, leading to severe indoor overheating degrees (IOD) and high carbon dioxide (CO2) concentrations if not adequately ventilated. To assess classroom design alternatives in the design development phase and retrofit scenarios, building energy simulation is a widely used method to estimate performance indicators. However, consideration of a high number of design alternatives increases computational cost and requires tedious modeling efforts. Research in building performance predictions with machine learning methods gained increasing interest in recent years. Artificial neural networks (ANNs) are reported to yield satisfactory performance in the prediction of non-linear patterns of building performance. This study presents a data-driven framework to estimate heating energy demand, IOD, and CO2 concentration of naturally ventilated classrooms with ANNs. Five input variables are selected to predict specified performance indicators. 200 classrooms with varying orientations, values of shape factor, glazing area, occupant density, and outdoor surface area are simulated. The ANNs are trained with a subset of EnergyPlus simulation results. Prediction models for three performance indicators are individually built, and prediction performances are evaluated. While regression coefficients range between 0.986 and 0.993, the average root mean square error calculated is between %2 and 9%, implying high predictive capacity.

Subject Keywords

Artificial neural networks, Building performance prediction, Building energy simulation

URI

https://hdl.handle.net/11511/97303

Collections

Department of Architecture, Conference / Seminar