TR2022-144
Extremum seeking controller tuning for heat pump optimization using failure-robust Bayesian optimization
-
- "Extremum seeking controller tuning for heat pump optimization using failure-robust Bayesian optimization", Journal of Process Control, DOI: 10.1016/j.jprocont.2022.11.006, Vol. 120, pp. 86-96, November 2022.BibTeX TR2022-144 PDF
- @article{Chakrabarty2022nov2,
- author = {Chakrabarty, Ankush and Burns, Daniel J. and Guay, Martin and Laughman, Christopher R.},
- title = {Extremum seeking controller tuning for heat pump optimization using failure-robust Bayesian optimization},
- journal = {Journal of Process Control},
- year = 2022,
- volume = 120,
- pages = {86--96},
- month = nov,
- doi = {10.1016/j.jprocont.2022.11.006},
- url = {https://www.merl.com/publications/TR2022-144}
- }
,
- "Extremum seeking controller tuning for heat pump optimization using failure-robust Bayesian optimization", Journal of Process Control, DOI: 10.1016/j.jprocont.2022.11.006, Vol. 120, pp. 86-96, November 2022.
-
MERL Contacts:
-
Research Areas:
Abstract:
Extremum seeking controllers have been investigated for multivariable data-driven energy optimization in heat pumps. In particular, proportional-integral extremum seeking control (PI-ESC) has demonstrated potential for significant accel- eration compared to other ESC variants for nonlinear closed-loop control systems. A barrier to PI-ESC’s utilization in self-optimizing control is the fact that the PI-ESC algorithm is fragile. That is, unless the PI-ESC gains (e.g., controller gains, estimator gains) are carefully tuned, small perturbations to these gains can render the closed-loop unstable. Since arbitrary combinations of PI-ESC gains can result in instabilities, we propose a failure-robust Bayesian optimization (FRBO) algorithm that computes PI-ESC gains that ensure the closed-loop system can be driven rapidly to the optimum, while identifying and avoiding regions in the space of PI-ESC gains that are likely to result in instabilities (i.e., failures). The FRBO-tuned PI-ESC is shown to result in rapid closed-loop convergence to optimal values both on benchmark examples and a production-level model of an air conditioning system.
Related Publication
- @inproceedings{Chakrabarty2022jul,
- author = {Chakrabarty, Ankush and Burns, Daniel J. and Guay, Martin and Laughman, Christopher R.},
- title = {Rapid Energy Optimization Of Vapor Compression Systems Using Probabilistic Machine Learning And Extremum Seeking Control},
- booktitle = {International Refrigeration and Air Conditioning Conference (IRACC)},
- year = 2022,
- month = jul,
- url = {https://www.merl.com/publications/TR2022-172}
- }