by Michael Maiworm, Daniel Limon, Jose Maria Manzano, Rolf Findeisen
Abstract:
We present an output feedback nonlinear model predictive control approach that uses a Gaussian process model for prediction. We show nominal stability assuming that the Gaussian process model is able to represent the real process and establish input-to-state stability assuming a bounded error between the real process and the Gaussian model approximation. These results are achieved using a predictive control formulation without terminal region. The approach is illustrated using a continuous stirred-tank reactor benchmark problem.
Reference:
Stability of Gaussian Process Learning Based Output Feedback Model Predictive Controls (Michael Maiworm, Daniel Limon, Jose Maria Manzano, Rolf Findeisen), In IFAC-PapersOnLine, volume 51, 2018.
Bibtex Entry:
@article{maiworm_stability_2018,
title = {Stability of {Gaussian} {Process} {Learning} {Based} {Output} {Feedback} {Model} {Predictive} {Controls}},
volume = {51},
issn = {2405-8963},
url = {http://www.sciencedirect.com/science/article/pii/S2405896318327058},
doi = {https://doi.org/10.1016/j.ifacol.2018.11.047},
abstract = {We present an output feedback nonlinear model predictive control approach that uses a Gaussian process model for prediction. We show nominal stability assuming that the Gaussian process model is able to represent the real process and establish input-to-state stability assuming a bounded error between the real process and the Gaussian model approximation. These results are achieved using a predictive control formulation without terminal region. The approach is illustrated using a continuous stirred-tank reactor benchmark problem.},
number = {20},
journal = {IFAC-PapersOnLine},
author = {Maiworm, Michael and Limon, Daniel and Manzano, Jose Maria and Findeisen, Rolf},
year = {2018},
keywords = {Gaussian processes, learning, output feedback, predictive control, robust, stability},
pages = {455 -- 461}
}