Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Evaluating data-driven building stock heat demand forecasting models for energy optimization

Tohid Jafarinejad, Ina De Jaeger, Arash Erfani, Dirk Saelens
Katholieke Universiteit Leuven, Belgium

Abstract: Achieving energy efficiency in building sector is plausible through overly non-convex energy optimization schemes such as demand side management (DSM) at district scale, which itself relies on the district’s thermal model. The main challenge is deriving accurate surrogate data-driven models for districts, which can compete with detailed time-intensive physics based models. In this study, common data-driven models namely ARX, ANN and SVM are deployed for modeling various districts under different aggregation criteria. This study is conducted based on the artificial districts that are developed for IBPSA Project 1. Furthermore, the datadriven models are compared based on two Key Performance Indicators (KPI) namely coefficient of determination (R2) and run-time. Finally, it is shown that an SVM model performs the best. Moreover, aggregating based on archetype yields more promising results in terms of accuracy.
Keywords: Building stock, Data driven model, Black box model, Identification
Pages: 310 - 317