Data-Driven Robust Finite-Iteration Learning Control for MIMO Nonrepetitive Uncertain Systems.

Area of Science:

• Control Systems Engineering
• Robotics
• Machine Learning

Background:

• Addressing challenges in fast finite-iteration convergence (FIC), nonrepetitive uncertainty, and data-driven design.
• Existing control methods struggle with systems exhibiting both nonrepetitive uncertainty and the need for rapid, predictable convergence.

Purpose of the Study:

• To propose a novel data-driven robust finite-iteration learning control (DDRFILC) for multiple-input-multiple-output (MIMO) nonrepetitive uncertain systems.
• To enable predetermination of both tracking error bounds and the number of iterations for convergence.

Main Methods:

• Reformulating the MIMO uncertain system into an iterative incremental linear model using a pseudo partitioned Jacobian matrix (PPJM).
• Estimating the PPJM iteratively via a projection algorithm and incorporating its estimation and error bounds into linear matrix inequalities (LMIs).
• Tuning the learning gain through the solution of LMIs to ensure bounded convergence within finite iterations.

Main Results:

• The proposed DDRFILC guarantees bounded convergence within a prespecified finite number of iterations.
• The method ensures that both the tracking error bound and the convergence iteration number can be designated beforehand.
• The approach effectively restrains the effects of PPJM estimation and its error bound on control performance.

Conclusions:

• The developed DDRFILC algorithm successfully achieves fast finite-iteration convergence for nonrepetitive uncertain MIMO systems.
• The data-driven approach offers precise control over convergence parameters, enhancing predictability and performance.
• Simulation results validate the efficacy and robustness of the proposed DDRFILC algorithm.