Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA

     

A machine learning-based methodology for harnessing the energy flexibility potential of residential buildings

Adamantios Bampoulas 1,2, Fabiano Pallonetto 1, Eleni Mangina 1,3, Donal P. Finn 1,2
1 UCD Energy Institute, University College Dublin, Dublin, Ireland
2 UCD School of Mechanical and Materials Engineering, University College Dublin, Dublin, Ireland
3 UCD School of Computer Science, University College Dublin, Dublin, Ireland


DOI: https://doi.org/10.26868/25222708.2021.30420
Abstract: A key issue in energy flexibility assessment is the lack of a scalable and end-user tailored approach to assess the flexibility of residential buildings. In this study, this problem is addressed by developing daily updated datadriven models of the zone temperature and the heating system power consumption based on dynamic feature selection and a sliding window method. The regression techniques used are random forests, neural networks, and support vector machines. The proposed methodology utilises synthetic data obtained from a calibrated whitebox model of a residential building for two indicative occupancy profiles. This research is likely to be of benefit to electricity system stakeholders to conduct short-term predictions of various target variables associated with building operation, and ultimately, to facilitate the evaluation of the flexibility potential of residential buildings in an end-user tailored manner. Results show that random forests combined with a sequential forward feature selection method exhibit the optimal performance both for the zone temperature and the heating load prediction models.
Keywords: energy flexibility, machine learning, data-driven, residential building, smart grid
Pages: 215 - 222
Paper:
bs2021_30420