Design and Analysis of a Self-Adaptive Zeroing Neural Network for Solving Time-Varying Quadratic Programming

Area of Science:

• Computational Intelligence
• Neural Networks
• Optimization Algorithms

Background:

• Time-varying quadratic programming (TVQP) problems require efficient and rapid solution methods.
• Traditional zeroing neural network (ZNN) models face limitations in convergence speed and adaptability.
• Fuzzy logic systems offer potential for enhancing neural network performance through self-adaptation.

Purpose of the Study:

• To develop a novel self-adaptive ZNN model for enhanced TVQP problem-solving.
• To integrate Takagi-Sugeno fuzzy logic system (TSFLS) for adaptive control within the ZNN framework.
• To investigate the convergence properties (finite-time and predefined-time) of the proposed model.

Main Methods:

• Design of a Takagi-Sugeno (T-S) fuzzy ZNN (TSFZNN) model incorporating a multiple-input-single-output TSFLS.
• Development and application of four novel activation functions: power-bi-sign (PBSAF), tanh-bi-sign (TBSAF), exp-bi-sign (EBSAF), and sinh-bi-sign (SBSAF).
• Theoretical analysis and experimental simulations to validate convergence performance.

Main Results:

• The TSFZNN model demonstrates superior convergence performance compared to traditional ZNN models.
• TSFZNN models utilizing PBSAF or TBSAF achieve finite-time convergence for TVQP problems.
• TSFZNN models employing EBSAF or SBSAF achieve predefined-time convergence for TVQP problems.

Conclusions:

• The proposed TSFZNN model effectively solves TVQP problems with enhanced convergence characteristics.
• The integration of TSFLS provides a robust self-adaptive mechanism for the ZNN model.
• The novel activation functions enable precise control over convergence time, offering flexibility in applications.