Data-efficient global extremum seeking framework for constrained optimization of nonlinear dynamical systems

Read full text Summarise with VIRAT
Authors: Wouter Weekers, Alessandro Saccon, Nathan van de Wouw
Cited by: 0
References: 35

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We present a novel, data-efficient, global extremum-seeking control (ESC) framework for performance optimization of dynamical systems with constraints that are analytically unknown but measurable. This framework extends existing global ESC approaches to (1) facilitate constrained optimization, and (2) reduce the number of measurements required to optimize performance. The proposed approach guarantees convergence to a neighborhood of the constrained global optimum for suitably tuned design variables; even in the presence of suboptimal local solutions or disconnected feasible sets. Moreover, performing measurements to gather the data required for ESC updates can be costly and/or time-consuming in practice, making it desirable to reduce the number of measurements required to optimize performance, even more so in a global optimization setting. Therefore, the proposed approach uses the collected data to construct online approximations of the cost and constraint functions using kernel-based function approximation methods. We provide verifiable bounds on the approximation errors of these kernel-based approximations. These bounds can subsequently be used to decide when to exploit the kernel-based approximations instead of performing a measurement. Herewith, the total number of measurements required to optimize performance can be reduced significantly. As such, the proposed approach makes more efficient use of previously collected data than existing ESC approaches. We show that the output of the optimization algorithm is guaranteed to converge to an arbitrarily small neighborhood of the set of global solutions of the constrained optimization problem. Consequently, potential constraint violations can be made arbitrarily small as well. Furthermore, the ability of the proposed approach to solve the constrained optimization problem, and the benefits of using the kernel-based approximations, are illustrated with a simulation example. In this simulation example, the use of the kernel-based approximations allows a reduction of more than 60% in the number of measurements required to converge to a neighborhood of the constrained global optimum compared to only using ‘regular’ ESC updates.

Fields of Study & Subjects

Extremum Seeking Control Systems Advanced Control Systems Optimization Metaheuristic Optimization Algorithms Research

Publication details

Journal / source
Automatica
Volume
188
Pages
112926–112926
Publication date
23 March 2026
Language
EN