Showing 1–1 of 1 results for author: Nobar, M
-
Guided Bayesian Optimization: Data-Efficient Controller Tuning with Digital Twin
Authors:
Mahdi Nobar,
Jürg Keller,
Alisa Rupenyan,
Mohammad Khosravi,
John Lygeros
Abstract:
This article presents the guided Bayesian optimization algorithm as an efficient data-driven method for iteratively tuning closed-loop controller parameters using an event-triggered digital twin of the system based on available closed-loop data. We define a controller tuning framework independent of the controller or the plant structure. Our proposed methodology is model-free, making it suitable f…
▽ More
This article presents the guided Bayesian optimization algorithm as an efficient data-driven method for iteratively tuning closed-loop controller parameters using an event-triggered digital twin of the system based on available closed-loop data. We define a controller tuning framework independent of the controller or the plant structure. Our proposed methodology is model-free, making it suitable for nonlinear and unmodelled plants with measurement noise. The objective function consists of performance metrics modeled by Gaussian processes. We utilize the available information in the closed-loop system to identify and progressively maintain a digital twin that guides the optimizer, improving the data efficiency of our method. Switching the digital twin on and off is triggered by data-driven criteria related to the digital twin's uncertainty estimations in the BO tuning framework. Effectively, it replaces much of the exploration of the real system with exploration performed on the digital twin. We analyze the properties of our method in simulation and demonstrate its performance on two real closed-loop systems with different plant and controller structures. The experimental results show that our method requires fewer experiments on the physical plant than Bayesian optimization to find the optimal controller parameters.
△ Less
Submitted 25 March, 2024;
originally announced March 2024.
