A novel predefined-time noise-tolerant zeroing neural network for solving time-varying generalized linear matrix equations

Abstract

Due to the time-varying nature of time-varying linear systems, many traditional time-invariant methods cannot solve time-varying linear systems. The emergence of zeroing neural networks (ZNN) has led to a long and intensive development in this field. Researchers have mainly studied or improved the ZNN model for time-varying linear systems in the direction of activation functions and design parameters. This paper proposes a novel modified versatile activation function (MVAF) from the perspective of activation function optimization, and a modified versatile activation function zeroing neural network (MVAFZNN) model for solving time-varying generalized linear matrix equations is constructed. The proposed model converges faster than the versatile activation function zeroing neural network (VAFZNN) model and other ZNN models. Theoretical analysis shows that the MVAFZNN model can solve time-varying systems in a predetermined time. Two numerical experiments verify the validity of the proposed model, and two circuit experiments show that the proposed model can be used to calculate circuit parameters.