Reinforcement Learning Optimal Control with A Model Identifier Based on Online Spectral Adaptive Law

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper proposes an reinforcement learning (RL) optimal control method based on an online spectral adaptive law (SPAL) model identifier. Distinguished from previous studies, to enhance the generalization capability of the neural network (NN) and the robustness of the system, the proposed method employs a rectified linear unit (ReLU) activated NN in the online model identifier, and utilizes a projection algorithm with a spectral adaptive law to update the NN weights online. By constructing a Lyapunov function incorporating the spectral norm of the NN weights, it is proven that the weights can converge to their true values.

Original languageEnglish
Title of host publicationProceedings of the 44th Chinese Control Conference, CCC 2025
EditorsJian Sun, Hongpeng Yin
PublisherIEEE Computer Society
Pages2711-2716
Number of pages6
ISBN (Electronic)9789887581611
DOIs
Publication statusPublished - 2025
Externally publishedYes
Event44th Chinese Control Conference, CCC 2025 - Chongqing, China
Duration: 28 Jul 202530 Jul 2025

Publication series

NameChinese Control Conference, CCC
ISSN (Print)1934-1768
ISSN (Electronic)2161-2927

Conference

Conference44th Chinese Control Conference, CCC 2025
Country/TerritoryChina
CityChongqing
Period28/07/2530/07/25

Keywords

  • Optimal control
  • Reinforcement learning
  • Spectral adaptive law

Fingerprint

Dive into the research topics of 'Reinforcement Learning Optimal Control with A Model Identifier Based on Online Spectral Adaptive Law'. Together they form a unique fingerprint.

Cite this