Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Feature selection and Gaussian Process regression for personalized thermal comfort prediction
Abstract User comfort is one of the main goals for the design of heating, ventilation and air conditioning systems. Therefore, well predicting comfort models are essential for keeping a room in comfortable conditions and furthermore enabling control approaches in accordance with occupants' preferences. Most comfort models target the representation of an average sensation of all occupants. However, in large office spaces, the thermal sensation can vary significantly between the occupants and averaged comfort predictions do not necessarily represent the individual preferences. To allow for customized thermal conditions, user feedback is collected in daily working routine and used for the development of a personalized comfort prediction model. Furthermore, the presented approach takes into account the effect of increased air movement to enable customized conditions. Individual comfort models are derived for each user based on a hybrid approach: The widely used predicted mean vote (PMV) is used as a starting point and evaluated for the given voting data. Based on this analysis, the proposed comfort model is composed of a polynomial basis function extended by a Gaussian Process regression model to capture the complex and highly subjective relations between room conditions and thermal perception. An analysis of the significance of all measured variables is performed to reduce the model's complexity by selecting the most impacting parameters. This approach allows for a 74% higher individual prediction accuracy compared to the standard PMV calculation.
Highlights Design of personal comfort models based on easily accessible data & user feedback Significance analysis of environmental conditions on individual comfort Regression analysis for feature selection using different model structures Consideration of transient conditions based on varying sampling times Implicit consideration of elevated air movement during the modeling process
Feature selection and Gaussian Process regression for personalized thermal comfort prediction
Abstract User comfort is one of the main goals for the design of heating, ventilation and air conditioning systems. Therefore, well predicting comfort models are essential for keeping a room in comfortable conditions and furthermore enabling control approaches in accordance with occupants' preferences. Most comfort models target the representation of an average sensation of all occupants. However, in large office spaces, the thermal sensation can vary significantly between the occupants and averaged comfort predictions do not necessarily represent the individual preferences. To allow for customized thermal conditions, user feedback is collected in daily working routine and used for the development of a personalized comfort prediction model. Furthermore, the presented approach takes into account the effect of increased air movement to enable customized conditions. Individual comfort models are derived for each user based on a hybrid approach: The widely used predicted mean vote (PMV) is used as a starting point and evaluated for the given voting data. Based on this analysis, the proposed comfort model is composed of a polynomial basis function extended by a Gaussian Process regression model to capture the complex and highly subjective relations between room conditions and thermal perception. An analysis of the significance of all measured variables is performed to reduce the model's complexity by selecting the most impacting parameters. This approach allows for a 74% higher individual prediction accuracy compared to the standard PMV calculation.
Highlights Design of personal comfort models based on easily accessible data & user feedback Significance analysis of environmental conditions on individual comfort Regression analysis for feature selection using different model structures Consideration of transient conditions based on varying sampling times Implicit consideration of elevated air movement during the modeling process
Feature selection and Gaussian Process regression for personalized thermal comfort prediction
Guenther, Janine (Autor:in) / Sawodny, Oliver (Autor:in)
Building and Environment ; 148 ; 448-458
15.11.2018
11 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Feature selection and Gaussian Process regression for personalized thermal comfort prediction
| British Library Online Contents | 2019
Assessment of Thermal Comfort Using Personalized Ventilation
| Taylor & Francis Verlag | 2010
The Personalized Thermal Comfort Prediction Using an MH-LSTM Neural Network Method
| DOAJ | 2024
| Taylor & Francis Verlag | 2021