An Accelerated Augmented Gradient Neural Network for Constrained Time-Varying Nonlinear Optimization

Area of Science:

• Optimization
• Neural Networks
• Control Systems

Background:

• Constrained time-varying nonlinear optimization (CTVNO) problems are critical in engineering.
• Existing models face challenges with complex architectures and slow convergence.
• These limitations hinder practical application in real-world scenarios.

Purpose of the Study:

• To introduce an Accelerated Augmented Gradient Neural Network (AAGNN) for CTVNO problems.
• To overcome the limitations of existing models, specifically complexity and slow response.
• To enhance the precision and speed of solving CTVNO problems.

Main Methods:

• Developed an AAGNN by integrating error and energy function gradients.
• Utilized theoretical analysis to demonstrate convergence properties.
• Conducted comparative numerical experiments on benchmark problems.

Main Results:

• The AAGNN demonstrated high-precision solutions with exponential error elimination.
• Achieved significantly lower residual errors compared to existing models.
• Showcased faster convergence rates in numerical experiments.

Conclusions:

• The AAGNN effectively reduces complexity and accelerates convergence for CTVNO.
• Validated superior performance and practical applicability through robotic control and portfolio selection.
• The proposed AAGNN offers a robust solution for complex optimization tasks.