Adaptive Optimal Control for Unknown Constrained Nonlinear Systems With a Novel Quasi-Model Network.

Area of Science:

• Control Theory
• Artificial Intelligence
• Nonlinear System Dynamics

Background:

• Optimal control of unknown continuous-time nonlinear systems is challenging.
• Adaptive dynamic programming (ADP) offers a framework for such problems.
• Traditional methods often require prior system model knowledge.

Purpose of the Study:

• To develop a policy-iteration-based algorithm for optimal control of unknown continuous-time nonlinear systems.
• To utilize adaptive dynamic programming (ADP) with neural networks.
• To eliminate the need for separate system model learning.

Main Methods:

• A policy-iteration-based algorithm using three neural networks (critic, actor, quasi-model).
• Simultaneous tuning of critic and quasi-model network parameters using the least sum of squares method.
• Iterative improvement of the control law based on optimality conditions.

Main Results:

• The algorithm effectively approximates the control law, cost function, and system dynamics.
• Simultaneous learning eliminates the need for pre-identified system models.
• Optimality and convergence properties of the proposed algorithm are demonstrated.

Conclusions:

• The presented adaptive dynamic programming algorithm is effective for optimal control of unknown nonlinear systems.
• The use of neural networks simplifies the control design process.
• Simulation results validate the algorithm's practical applicability.