An adaptive variable-parameter dynamic learning network for solving constrained time-varying QP problem.

Area of Science:

• * Computational Mathematics
• * Neural Networks
• * Optimization Theory

Background:

• * Existing methods for time-varying convex quadratic programming (TVCQP) include varying-parameter and fixed-parameter convergent-differential neural networks (VPCDNN and FPCDNN).
• * These methods have limitations in convergence speed and error bounds for solving dynamic optimization problems.

Purpose of the Study:

• * To propose and analyze a novel adaptive variable-parameter dynamic learning network (AVDLN) for efficiently solving TVCQP problems with equational constraints.
• * To demonstrate the superiority of AVDLN over existing VPCDNN and FPCDNN methods in terms of convergence and robustness.

Main Methods:

• * Transforming the TVCQP problem into a time-varying matrix equation.
• * Designing an adaptive time-varying formulation for the error function and integrating it into the time-varying parameter.
• * Proving convergence and robustness theorems using Lyapunov stability analysis.

Main Results:

• * Mathematical analysis shows AVDLN has a smaller upper bound on convergence error and a faster error convergence rate than FPCDNN and VPCDNN.
• * Simulations confirm AVDLN's validity, demonstrating a faster convergence speed and smaller error fluctuation.

Conclusions:

• * The proposed AVDLN is an effective and efficient method for solving TVCQP problems.
• * AVDLN offers significant improvements in convergence speed and error reduction compared to existing neural network approaches.