Virtual Target-Oriented Neural Learning for Robust Optimal Tracking Control of Discrete Strict-Feedback Systems.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Nonlinear strict-feedback systems (SFSs) present significant control challenges due to unknown dynamics and unmatched disturbances (uMDs).
• Traditional discrete-time control methods for SFSs often suffer from noncausality issues.
• Model-based control requires accurate system identification, which is often infeasible.

Purpose of the Study:

• To develop a data-driven hierarchical neural learning (HNL) algorithm for optimal tracking control (OTC) of nonlinear SFSs.
• To address challenges posed by unmatched disturbances (uMDs) and unknown system dynamics.
• To eliminate noncausal issues in discrete-time SFS control.

Main Methods:

• Proposed a virtual target (VT) construction scheme leveraging the recursive structure of SFSs.
• Employed a time-varying affine Hamilton-Jacobi-Isaacs (HJI) formulation to link auxiliary control and disturbance.
• Utilized an adaptive dynamic programming (ADP) framework with a novel tracking network (T-network) for controller synthesis from input-output data.

Main Results:

• The HNL algorithm achieved optimal tracking control for nonlinear SFSs without requiring an accurate plant model.
• The proposed T-network enhanced policy updates by merging gradient information and future tracking errors.
• Simulations demonstrated outstanding performance, robustness to uMDs, and tolerance to significant model uncertainties.

Conclusions:

• The developed HNL algorithm offers a robust and effective solution for OTC of complex nonlinear systems.
• The data-driven approach eliminates the need for precise system models, making it broadly applicable.
• The enhanced neural architecture ensures simultaneous reduction in control effort and improvement in tracking accuracy.