Dealing With the Issues Crucially Related to the Functionality and Reliability of NN-Associated Control for Nonlinear...

Area of Science:

• Control Engineering
• Artificial Intelligence
• Machine Learning

Background:

• The universal approximation/learning feature of neural networks (NNs) is crucial for control design.
• This feature is contingent upon critical conditions that are often overlooked in existing NN-based control designs.
• Unsatisfied conditions can render the NN's learning capability ineffective.

Purpose of the Study:

• To identify and address fundamental issues in NN-based control design.
• To propose a novel strategy for ensuring NNs are fully functional in control loops.
• To enhance the reliability and effectiveness of NN-associated control performance.

Main Methods:

• Introduction of a new structural NN unit with diversified neurons and self-adjusting subneurons.
• Ensuring input signals to subneurons are confined within a compact set.
• Guaranteeing continuity of control signals and boundedness of all closed-loop signals.

Main Results:

• The proposed method ensures the NN unit remains fully functional throughout system operation.
• Reliable and effective NN-associated control performance is achieved.
• Theoretical analysis and numerical simulations validate the proposed approach.

Conclusions:

• The study successfully addresses critical overlooked issues in NN-based control design.
• The novel NN structure enhances the robustness and effectiveness of control systems.
• The findings contribute to more dependable applications of NNs in control engineering.