Scholarlink

<p>This article addresses robust tracking control of linear parameter varying systems subject to time-varying external disturbances and parametric uncertainties, where conventional fixed-gain disturbance observers and memoryless neural observers suffer from a bandwidth–frequency mismatch and from a lack of temporal memory. An interpretable disturbance observer-based control framework is proposed, in which the observer gain is generated by closed-form continuous-time neural dynamics and projected onto a parameter-dependent stability-ensuring set defined via a linear matrix inequality. The composite control law combines a state-dependent Riccati equation feedback term with an observer-based feedforward compensation. A Lyapunov-based analysis establishes uniform ultimate boundedness of the disturbance estimation error and input-to-state stability of the closed loop, including explicit treatment of parameter rate of variation, unmatched components, measurement noise, and online learning. Simulation results on a linear parameter varying model of a morphing aircraft, supported by a Monte Carlo study with thirty independent trials under additive sensor noise and varying scheduling profiles, demonstrate superior tracking accuracy and disturbance estimation quality of the proposed observer compared to a multilayer perceptron based disturbance observer and a fixed-gain disturbance observer, while keeping the computational burden compatible with real-time implementation. The novelty lies in: (i) the use of closed-form continuous-time neural dynamics to generate a time-varying, stability-preserving observer gain that is natively robust to discretisation; (ii) a rigorous stability analysis combining a non-expansive eigenvalue projection with a parameter-dependent Riccati synthesis; (iii) an interpretable observer whose internal neuronal activities can be inspected to relate disturbance episodes to gain adaptation; and (iv) A channel-aware constraint on the CfC readout, with smoothing on the unmatched part, so the residual error stays consistent with the model’s theoretical limit.</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

Interpretable CfC-DOBC of Robust Tracking Control for a Class of Linear Parameter Varying Systems

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Christian-Marie Moanda Ndeko Mosengo,
  • Hongwei Mo,
  • Sazedul Haque,
  • Salaudden Usuf

摘要

This article addresses robust tracking control of linear parameter varying systems subject to time-varying external disturbances and parametric uncertainties, where conventional fixed-gain disturbance observers and memoryless neural observers suffer from a bandwidth–frequency mismatch and from a lack of temporal memory. An interpretable disturbance observer-based control framework is proposed, in which the observer gain is generated by closed-form continuous-time neural dynamics and projected onto a parameter-dependent stability-ensuring set defined via a linear matrix inequality. The composite control law combines a state-dependent Riccati equation feedback term with an observer-based feedforward compensation. A Lyapunov-based analysis establishes uniform ultimate boundedness of the disturbance estimation error and input-to-state stability of the closed loop, including explicit treatment of parameter rate of variation, unmatched components, measurement noise, and online learning. Simulation results on a linear parameter varying model of a morphing aircraft, supported by a Monte Carlo study with thirty independent trials under additive sensor noise and varying scheduling profiles, demonstrate superior tracking accuracy and disturbance estimation quality of the proposed observer compared to a multilayer perceptron based disturbance observer and a fixed-gain disturbance observer, while keeping the computational burden compatible with real-time implementation. The novelty lies in: (i) the use of closed-form continuous-time neural dynamics to generate a time-varying, stability-preserving observer gain that is natively robust to discretisation; (ii) a rigorous stability analysis combining a non-expansive eigenvalue projection with a parameter-dependent Riccati synthesis; (iii) an interpretable observer whose internal neuronal activities can be inspected to relate disturbance episodes to gain adaptation; and (iv) A channel-aware constraint on the CfC readout, with smoothing on the unmatched part, so the residual error stays consistent with the model’s theoretical limit.