State Estimation For Discrete Time-varying Neural Networks Under Communication Protocol

Artificial neural network is a mathematical model which simulates the behavioral characteristics of human brain and animal neural networks for distributed and parallel information processing.In this case,the neural network is very complicated.What's more,the neural network has the strong ability of self-learning and self-adaptation.Furthermore,neural network adjust the connection mode between internal neurons to realize the information exchange processing.The neural network has a great deal of neurons and the structure is complex,it is difficult to measure its state,so we need to use the available measurement to estimate state.In addition,during the communication network environment,data transmission between nodes may inevitably occur data collisions,data congestion,data delays and other phenomena due to communication constraints.Therefore,communication protocols are need to introduce to assign nodes access to communication channels.The main research contents of this paper are as below:Firstly,the state estimation approach is investigated for a kind of neural networks with state saturation and packet loss based on Round-Robin(RR)protocol.Uncorrelated random variables that obey the Bernoulli distribution are used to describe the randomly occurring nonlinear activation function.The state estimator is designed by using the measurement information with data packet dropout,so that the error dynamic system meets the asymptotic stability in the mean square.Then,the estimator parameter is obtained by utilizing the linear matrix inequality method.Numerical simulation is used to show the availability of the designed state estimation method.Secondly,the problem of state estimation is explored for a class of discrete neural networks with time-varying delays and false data injection attacks under Weighted Try-Once-Discard(WTOD)protocol.The false data injection attacks are modelled by using the Bernoulli distributed random variable.The sensor nodes subject to the WTOD protocol which can transmit data selectively.The WTOD protocol can reduce unnecessary data transmission and alleviate the bandwidth pressure.Then,state estimator is designed based on the obtained measurement to effectively estimate the neurons' state.By utilizing Lyapunov stability theory,the sufficient conditions for error dynamic system are given to ensure the exponentially ultimate boundedness in the mean square.A numerical example is provided to testify the feasibility of the proposed state estimation approach.Thirdly,the H? state estimation problem is analyzed for a class of discrete time-varying neural networks with multirate sampling.In this case,we consider the phenomenon of fading channels in the process of signal transmission.What's more,a new measurement information model is established according to the SCP and a state estimator is designed.By means of the Lyapunov stability theory and random analysis method,some conditions which guarantee the stochastically stable of the estimation error system are gained.At the same time,by utilizing the linear matrix inequality technology,the explicit expression of the state estimator gain matrix is acquired.An illustrative example is applied to demonstrate the effectiveness of the estimation method.