Scholarlink

<p>This paper introduces a novel, scalable framework for synchronizing heterogeneous nonlinear multi-agent systems with distinct agent dynamics, integrating data-driven modeling, robust state estimation, and decentralized control. The approach comprises four key components: (1) Sparse Identification of Nonlinear Dynamics (SINDy) to identify unique dynamics for each agent from data, (2) feedback linearization to transform each agent’s nonlinear dynamics into a linear form, (3) a neural network-based extended state observer (NNESO) to estimate states and lumped uncertainty, unmodeled dynamics and external disturbances, and (4) decentralized linear model predictive control(MPC) for leader-follower synchronization. In contrast to traditional nonlinear MPC, which faces computational challenges in heterogeneous systems with high nonlinearity or large agent populations, this framework leverages SINDy and feedback linearization to enable tractable, real-time control, while the NNESO ensures robustness to modeling errors and inter-agent variations, enabling the use of a tractable, decentralized linear MPC. The approach is validated through case studies on cooperative stabilization of a heterogeneous multi-agent system and synchronization of a network of Van der Pol oscillators with distinct dynamics. Comparative evaluations against nonlinear MPC and neural-network-based sliding mode control demonstrate superior tracking performance and up to 80% reduction in online computation time, underscoring the framework’s effectiveness and scalability for complex multi-agent systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust data-driven synchronization of heterogeneous nonlinear multi-agent systems

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Shahin Razani,
  • Mahsan Tavakoli-Kakhki,
  • Ahmad Kalhor

摘要

This paper introduces a novel, scalable framework for synchronizing heterogeneous nonlinear multi-agent systems with distinct agent dynamics, integrating data-driven modeling, robust state estimation, and decentralized control. The approach comprises four key components: (1) Sparse Identification of Nonlinear Dynamics (SINDy) to identify unique dynamics for each agent from data, (2) feedback linearization to transform each agent’s nonlinear dynamics into a linear form, (3) a neural network-based extended state observer (NNESO) to estimate states and lumped uncertainty, unmodeled dynamics and external disturbances, and (4) decentralized linear model predictive control(MPC) for leader-follower synchronization. In contrast to traditional nonlinear MPC, which faces computational challenges in heterogeneous systems with high nonlinearity or large agent populations, this framework leverages SINDy and feedback linearization to enable tractable, real-time control, while the NNESO ensures robustness to modeling errors and inter-agent variations, enabling the use of a tractable, decentralized linear MPC. The approach is validated through case studies on cooperative stabilization of a heterogeneous multi-agent system and synchronization of a network of Van der Pol oscillators with distinct dynamics. Comparative evaluations against nonlinear MPC and neural-network-based sliding mode control demonstrate superior tracking performance and up to 80% reduction in online computation time, underscoring the framework’s effectiveness and scalability for complex multi-agent systems.