Adaptive Variable Parameter ZNN for Solving Time-varying Linear Matrix Equation

Abstract

An adaptive fuzzy variable parametric zeroing neural network (AFV-ZNN) model is proposed to solve time-varying linear matrix equation. Distinguishing from varying-gain recurrent neural network models and other zeroing neural network models, the fuzzy design parameter of AFV-ZNN is adaptive as based on the residual matrix. The model adopts three different activation functions to achieve a faster convergence rate. The theoretical analysis and examples illustrate the faster convergence of the ATV-ZNN model, and the better performance of activation functions, compared with existing neural networks in simulation experiments.