Accelerated Gradient Approach For Neural Network Adaptive Control Of Nonlinear Systems

Recent connections in the adaptive control literature to continuous-time analogues of Nesterov's accelerated gradient method have led to the development of new real-time adaptation laws based on accelerated gradient methods. However, previous results assume the system's uncertainties are l...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Proceedings of the IEEE Conference on Decision & Control s. 3475 - 3480
Hlavní autoři: Le, Duc M., Patil, Omkar Sudhir, Nino, Cristian F., Dixon, Warren E.
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 06.12.2022
Témata:
ISSN:2576-2370
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Recent connections in the adaptive control literature to continuous-time analogues of Nesterov's accelerated gradient method have led to the development of new real-time adaptation laws based on accelerated gradient methods. However, previous results assume the system's uncertainties are linear-in-the-parameters (LIP). In this paper, a new NN-based accelerated gradient adaptive controller is developed to achieve trajectory tracking in general nonlinear systems subject to unstructured uncertainties that do not satisfy the LIP assumption. Higher-order accelerated gradient-based adaptation laws are developed to generate real-time estimates of both the unknown ideal output-and hidden-layer weights of a NN. A nonsmooth Lyapunov-based method is used to guarantee the closed-loop error system achieves global asymptotic tracking. Simulations are conducted to demonstrate the improved performance from the developed method. Results show the higher-order adaptation outperforms the standard gradient-based NN adaptation by 32.3% in terms of the root mean squared function approximation error.
ISSN:2576-2370
DOI:10.1109/CDC51059.2022.9993122