Skip to Main content Skip to Navigation

Data-driven building thermal modeling using system identification for hybrid systems

Abstract : The building sector is a major energy consumer, therefore, a framework of actions has been decided on by countries worldwide to limit its impact. For implementing such actions, the availability of models providing an accurate description of the thermal behavior of buildings is essential. For this purpose, this thesis proposes the application of a new data-driven technique for modeling the thermal behavior of buildings based on a hybrid system approach. Hybrid systems exhibit both continuous and discrete dynamics. This choice is motivated by the fact that a building is a complex system characterized by nonlinear phenomena and the occurrence of different events. We use a PieceWise AutoRegressive eXogeneous inputs (PWARX) model for the identification of hybrid systems. It is a collection of sub-models where each sub-model is an ARX equation representing a certain configuration in the building characterized by its own dynamics. This thesis starts with a state-of-the-art on building thermal modeling. Then, the choice of a hybrid system approach is motivated by a mathematical interpretation based on the equations derived from an RC thermal circuit of a building zone. This is followed by a brief background about hybrid system identification and a detailed description of the PWARX methodology. For the prediction phase, it is shown how to use the Support Vector Machine (SVM) technique to classify new data to the right sub-model. Then, it is shown how to integrate these models in a hybrid control loop to estimate the gain in the energy performance for a building after insulation work. The performance of the proposed technique is validated using data collected from various test cases.
Document type :
Complete list of metadatas

Cited literature [195 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Wednesday, July 15, 2020 - 1:39:09 PM
Last modification on : Wednesday, August 5, 2020 - 3:49:41 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02899639, version 1


Balsam Ajib. Data-driven building thermal modeling using system identification for hybrid systems. Automatic. Ecole nationale supérieure Mines-Télécom Lille Douai, 2018. English. ⟨NNT : 2018MTLD0006⟩. ⟨tel-02899639⟩



Record views


Files downloads