A Novel Swarm-Exploring Neurodynamic Network for Obtaining Global Optimal Solutions to Nonconvex Nonlinear...

Area of Science:

• Computational Neuroscience
• Optimization Theory
• Artificial Intelligence

Background:

• Nonconvex nonlinear programming problems present significant challenges in various scientific and engineering fields.
• Existing methods, such as collaborative neurodynamic approaches (CNA), often require multiple recurrent neural network (RNN) runs, limiting efficiency.
• There is a need for more efficient and globally capable neurodynamic models for solving complex optimization tasks.

Purpose of the Study:

• To propose a novel swarm-exploring neurodynamic network (SENN) for solving nonconvex nonlinear programming problems.
• To enhance the global search capabilities of neurodynamic models.
• To demonstrate the efficiency and superiority of the proposed SENN compared to existing methods.

Main Methods:

• Development of a two-timescale convergent-differential (TTCD) model by integrating a convergent-differential neural network (CDNN) as a local quadratic programming (QP) solver with a two-timescale design.
• Incorporation of swarm exploration neurodynamics into the TTCD model to create the SENN, endowing it with global search capabilities.
• Stability analysis of the TTCD model and simulation-based demonstration of the SENN's feasibility and performance.

Main Results:

• The proposed SENN effectively solves nonconvex nonlinear programming problems.
• The SENN demonstrates superior performance compared to existing collaborative neurodynamic approaches (CNA).
• A key advantage of the SENN is its ability to achieve global search using a single recurrent neural network (RNN), unlike CNA which requires multiple RNN runs.

Conclusions:

• The developed swarm-exploring neurodynamic network (SENN) offers an effective and efficient solution for nonconvex nonlinear programming.
• The SENN's integration of swarm exploration with a two-timescale model provides enhanced global search capabilities.
• The SENN represents a significant advancement over existing collaborative neurodynamic methods due to its simplified architecture and improved performance.