Nonlinear Characteristic Analysis And Synchronization Control Method For Memristor Neural Networks

With the popularization and rapid development of artificial intelligence,neural networks have received more and more attention and played an important role in image processing,speech recognition,automatic control,transportation and other fields.In order to implement artificial intelligence applications,the traditional Von Neumann computers have been greatly challenged with data processing capabilities and command execution speed.Inspired by neural networks,the computing architectures have been evolving into an intelligent architecture that is different from the traditional Von Neumann computers.In 2008,HP researchers found that memristive characteristics existed in nano-scale RRAM,and announced that they had found a memristor proposed by Professor Leon O.Chua in 1971.A synapse works similarly to a memristor,and some memristive devices are also considered as good candidates for neural network applications.Compared with neural networks with fixed weights,the weight of memristive neural networks is variable,which improves the learning ability of neural networks and is closer to biological neural networks.The study on the synchronization control can help understand the dynamics of memristive neural networks and how these networks represent,process,and exchange information through the time evolution of their activities.Because memristive neural networks do not have a solid and mature theoretical system and there is a huge potential application value,this paper proposes the finite-time synchronization control of the memristive Cohen-Grossberg neural network with time-varying delays,the pinning synchronization control of the memristive Cohen-Grossberg neural network with mixed delays,and the stability of the freeways networks based on discrete cell neural networks.The main research results are as follows:(1)Based on the memristive Cohen-Grossberg neural networks with time-varying delays,two kinds of piecewise control are proposed,and the finite-time synchronization conditions for the two memristive Cohen-Grossberg neural networks with time-varying delays via the two piecewise controls are given.In addition,compared with the normal feedback control,the two piecewise controls can shorten the synchronization time.Due to the universality of CohenGrossberg neural networks,Hopfield neural networks can be obtained through simple deformation.Therefore,this main content can be applied to the memristive Hopfield neural networks with time-varying delays.In the same initial conditions,compared with normal feedback control,numerical simulations show that these two piecewise controllers can shorten the synchronization time.(2)In actual circuit,in addition to the signal delays caused by the amplifiers,there are also distribution delays caused by the structure of the neural network.In this paper,a hybrid control composed of pinning control and intermittent control is proposed.And exponential synchronization conditions for the memristive Cohen-Grossberg neural networks with timevarying delays and distributed delays are given.In addition to being applied to Hopfield neural networks,it can be also applied to a class of memristive recurrent neural networks with no delay,time-varying delays,distributed delays or mixed delays by changing the control of delay parameters.Numerical simulations show the feasibility of the hybrid control.(3)The freeway networks are a class of typical complex networks.In this paper,CTM is used to simplify the freeway networks into discrete cellular neural networks.According to the stability theorems,the existence of the equilibrium points of the freeway networks are proposed,and the conditions of the robust global exponential stability of the equilibrium point of the freeway network are given.Numerical simulations show that the freeway networks simplified as a discrete cell neural networks have equilibrium points which are robustly globally exponentially stable.