A neurodynamic approach to nonlinear optimization problems with affine equality and convex inequality constraints

Area of Science:

• Optimization
• Computational Neuroscience
• Applied Mathematics

Background:

• Nonlinear optimization problems are prevalent in science and engineering.
• Existing neural network approaches often require restrictive assumptions.
• Efficient and robust optimization methods are crucial for complex systems.

Purpose of the Study:

• To propose a new neurodynamic approach for nonlinear optimization with constraints.
• To ensure finite-time convergence and global optimality.
• To overcome limitations of existing neural network optimization methods.

Main Methods:

• Development of a novel neural network architecture.
• Incorporation of a time-varying auxiliary function.
• Analysis of convergence properties for nonconvex and convex objective functions.

Main Results:

• The proposed neural network guarantees finite-time entry into the feasible region.
• The network converges to the critical point set for generally nonconvex objectives.
• Global convergence to an optimal solution is proven for pseudoconvex/convex objectives.
• The method relaxes common restrictive assumptions on feasible regions and objective functions.

Conclusions:

• The neurodynamic approach offers a robust and efficient method for constrained nonlinear optimization.
• This method demonstrates superior convergence properties without stringent assumptions.
• The approach is validated through numerical examples and a real-time data reconciliation application.