A novel zeroing neural network model for the time-varying quadratic optimization problem

The time-varying quadratic optimization problem (TVQOP) is a theoretical foundation for industrial applications. Solving the TVQOP quickly and accurately is difficult, being limited by the complexity of the coefficient matrix dimensions. A novel zeroing neural network (NZNN) model is proposed, targeting the TVQOP with enhanced convergence efficiency. The NZNN model adopts a novel activation function (NAF). The NAF is designed to incorporate particular piecewise and nonlinear functions distributed in the real number field. The NZNN model with NAF achieves convergence in finite time and is robust to noises. The convergence properties and anti-interference capacity of the NZNN model are rigorously derived through Lyapunov theory. Comparative evaluations against zeroing neural network variants reveal its advantages, including high-precision solutions, accelerated convergence and sustained robustness under different disturbances.