Composite learning from adaptive backstepping neural network control.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Existing neural network (NN) learning control methods require persistent excitation (PE) for NN parameter convergence.
• This condition is often difficult to satisfy in practical applications of adaptive control for nonlinear systems.

Purpose of the Study:

• To develop a new NN composite learning technique for command-filtered backstepping adaptive control.
• To guarantee NN weight convergence without the need for persistent excitation in strict-feedback nonlinear systems with functional uncertainties.

Main Methods:

• A spatially localized NN approximation is used to address functional uncertainties.
• Online historical and instantaneous data are utilized to generate prediction errors.
• Both tracking errors and prediction errors are employed for updating NN weights.

Main Results:

• The proposed NN composite learning method guarantees NN parameter convergence without the persistent excitation condition.
• The influence of NN approximation errors on control performance is quantified.
• Effectiveness and superiority are demonstrated through illustrative results.

Conclusions:

• The developed NN composite learning technique offers a robust solution for adaptive control of nonlinear systems.
• It eliminates the stringent persistent excitation requirement, enhancing practical applicability.
• The method shows significant advantages over existing NN learning control approaches.