TY - CONF

T1 - Component-Based Machine Learning for HVAC Systems Component Modeling

AU - Nabavi, Seyed Azad

AU - Saluz, Ueli

AU - Mohammadi, Sahar

AU - Geyer, Philipp

PY - 2023

Y1 - 2023

N2 - Heating Ventilation and Air Conditioning (HVAC) systems are responsible for a significant portion of building energy consumption, accounting for up to 38% and 12% of global energy consumption. Predicting energy consumption for HVAC systems in the early design phases is important due to their significant impact on energy use and user comfort. However, it is a challenging task due to the complex and dynamic nature of these systems traditionally requiring the effort of building simulation. The main aim of this research is to use machine learning (ML) techniques to model the components of HVAC systems in buildings and to predict the system's performance. We analyze the HVAC components individually to assess the proposed component-based machine learning method's ability to predict their performance and explore their interdependence. The components are structured in two alternative hierarchies to examine alternative modeling approaches: the first hierarchy's order follows the direction of energy flows with the order Z-S-P (zone, secondary HVAC, and primary HVAC components), while the second one follows the logic of design and engineering with the order Z-P-S. A random forest regression algorithm serves as a component ML model. The R2 value for the CBML model is, respectively, 0.98, 0.99, and 0.99 in forecasting the zone, primary HVAC, and secondary HVAC components in the Z-P-S hierarchy. Hence, the component-based ML method is highly effective in forecasting HVAC system components especially, in the Z-P-S hierarchy. Moreover, in forecasting the secondary HVAC components, the hierarchy following the design and engineering logic shows a significantly higher accuracy for the heat transfer coefficient. The comparison of the prediction accuracy of the CBML method in both hierarchies highlights the critical role of design dependencies in defining such data-driven prediction hierarchies. The primary HVAC component configuration playing a crucial role in modeling secondary HVAC components is a representative example of such a situation.

AB - Heating Ventilation and Air Conditioning (HVAC) systems are responsible for a significant portion of building energy consumption, accounting for up to 38% and 12% of global energy consumption. Predicting energy consumption for HVAC systems in the early design phases is important due to their significant impact on energy use and user comfort. However, it is a challenging task due to the complex and dynamic nature of these systems traditionally requiring the effort of building simulation. The main aim of this research is to use machine learning (ML) techniques to model the components of HVAC systems in buildings and to predict the system's performance. We analyze the HVAC components individually to assess the proposed component-based machine learning method's ability to predict their performance and explore their interdependence. The components are structured in two alternative hierarchies to examine alternative modeling approaches: the first hierarchy's order follows the direction of energy flows with the order Z-S-P (zone, secondary HVAC, and primary HVAC components), while the second one follows the logic of design and engineering with the order Z-P-S. A random forest regression algorithm serves as a component ML model. The R2 value for the CBML model is, respectively, 0.98, 0.99, and 0.99 in forecasting the zone, primary HVAC, and secondary HVAC components in the Z-P-S hierarchy. Hence, the component-based ML method is highly effective in forecasting HVAC system components especially, in the Z-P-S hierarchy. Moreover, in forecasting the secondary HVAC components, the hierarchy following the design and engineering logic shows a significantly higher accuracy for the heat transfer coefficient. The comparison of the prediction accuracy of the CBML method in both hierarchies highlights the critical role of design dependencies in defining such data-driven prediction hierarchies. The primary HVAC component configuration playing a crucial role in modeling secondary HVAC components is a representative example of such a situation.

UR - http://www.scopus.com/inward/record.url?scp=85207200122&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85207200122

T2 - 30th International Conference on Intelligent Computing in Engineering 2023, EG-ICE 2023

Y2 - 4 July 2023 through 7 July 2023

ER -