Nonlinearity Activated Noise-Tolerant Zeroing Neural Network for Real-Time Varying Matrix Inversion

Abstract

Real-time varying matrix inversion is widely used in the fields of science and engineering, e.g., image processing, signal processing and robot technology, etc. In this paper, a nonlinearity activated noise-tolerant zeroing neural network (NANTZNN) is constructed and employed to the time-dependent matrix inversion in the noisy environment. Compared with the gradient approach related neural network (GNN) and the existing noise-tolerant zeroing neural network (NTZNN), the proposed NANTZNN model is activated by specially-constructed nonlinear activation functions, and thus possesses the better convergence performance. Additionally, theoretical analyses are provided to guarantee the convergence of the proposed model. Finally, simulations are conducted to demonstrate the efficiency and superiority of the NANTZNN model for time-dependent matrix inversion, as compared with the NTZNN model.