Dynamic Behavior Analysis And Application Of Several Kinds Of Neural Networks

Artificial neural network is a network structure with distributed and parallel information processing function obtained by researchers by abstractly modeling the human brain nervous system,and the information processing mechanism used to show the connections between neurons is also called(intelligence principle).According to the special correspondence between the electrical characteristics of the connections between neurons and the threshold,people use the basic circuit to implement the neural circuit architecture of the group of neurons.Because the traditional artificial neural network circuit structure cannot meet the technical requirements of the brain-like intelligent chip in terms of power consumption and circuit size.And researchers focused their attention on the emerging information storage and processing components of memristors.Based on the characteristics of fast power-off retention,continuously adjustable resistance,low power consumption,nanometer size,etc.Researchers focused on seen as a natural electronic neuron synapse.It further promotes the research progress of memristor-based artificial neural networks in applied research fields such as brain-like intelligence,associative memory,and nonlinear systems.It is also the theoretical analysis method of memristor-based neural network internal dynamic behavior analysis.At the same time,it also opens up a new challenge for the theoretical analysis method of internal dynamic behavior analysis of neural network based on memristor.The research contents and main results of this article are as follows:Firstly,the stability of the switching network based on the third-order memristive circuit is studied based on the nonlinear storage memory characteristics of the new type of memristor element in the traditional third-order Chua's circuit.In this section,the nonlinear output characteristics of the third-order memristive circuit model are analyzed first,and then the stability characteristics of the switching network based on the third-order memristive circuit are analyzed by combining the Lyapunov stability analysis method and Filippov's discontinuous Dini-derivative theory.The validity of the analysis method is verified by numerical simulation,and it provides a method reference for the theoretical research of memristive switching complex networks.Secondly,for a Hopfield neural network with delayed memristive switching control,set-valued mapping Filippov's discontinuous Dini-derivative theory and Mmatrix theory on the right are applied to the Hopfield neural network based on delayed memristive switching control.The dynamic behavior characteristics were systematically analyzed.By constructing an appropriate Lyapunov-Krasovskii functional equation and combining Lyapunov's second stability theorem,the theoretical analysis of the globally uniform asymptotic stability of the system's equilibrium point is performed.Finally,a numerical simulation case verifies the correctness and effectiveness of the analysis method.It also makes a theoretical extension for the analysis method of delayed feedback neural network based on memristance.Thirdly,memory,analysis,and association are the three main functions of the human brain.This section uses as a practical application case of convolutional neural networks in vehicle detection and model recognition to illustrate the investment in deep learning associative memory networks in a real physical environment.Terminal devices in commercial applications do not meet the issues of low power consumption,small size,and lack of computing performance.Furthermore,this paper expounds the research methods and ideas of scientists in the field of compilable AI computers based on memristors.