Journal
BUILDING SIMULATION
Volume 15, Issue 3, Pages 317-331Publisher
TSINGHUA UNIV PRESS
DOI: 10.1007/s12273-021-0811-x
Keywords
demand response; support vector regression; machine learning; building peak demand; model predictive control; smart grid
Categories
Funding
- National Natural Science Foundation of China [51908365, 71772125]
- Philosophical and Social Science Program of Guangdong Province [GD18YGL07]
Ask authors/readers for more resources
This study developed a data-driven model predictive control using support vector regression for fast demand response events in commercial buildings. By optimizing SVR hyperparameters and shortening the genetic algorithm search range, the proposed SVR-based MPC successfully achieved simultaneous control of power demand and indoor temperature. Compared with RC-based MPC, the SVR-based MPC reduced time/labor costs without sacrificing control performance in fast DR events.
Demand response (DR) of commercial buildings by directly shutting down part of operating chillers could provide an immediate power reduction for power grids. In this special fast DR event, effective control needs to guarantee expected power reduction and ensure an acceptable indoor environment. This study, therefore, developed a data-driven model predictive control (MPC) using support vector regression (SVR) for fast DR events. According to the characteristics of fast DR events, the optimized hyperparameters of SVR and shortened searching range of genetic algorithm are used to improve the control performance. Meanwhile, a comprehensive comparison with RC-based MPC is conducted based on three scenarios of power demand controls. Test results show that the proposed SVR-based MPC could fulfill the control objectives of power demand and indoor temperature simultaneously. Compared with RC-based MPC, the SVR-based MPC could alleviate the time/labor cost of model development without sacrificing the control performance of fast DR events.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available