Dynamic Event-triggered H_? State Estimation Of Discrete Time-delay Neural Networks

In the past decades,neural network is widely used in many fields and a large number of research results have been obtained.With the rapid development of information technology,the combination of intelligent network and neural network has made a breakthrough,but at the same time,it also brings new challenges to the research of neural network in the networked environment.In the networked environment,due to various reasons such as physical constraints and technical difficulties,the network state information is not always fully accessible in practice.In this case,it is particularly urgent to study the state estimation of neural networks.This thesis is concerned with the event-triggered H? state estimation problem for a class of discrete time-delayed neural networks.There are some network-induced phenomena in the neural network model we considered.Firstly,the time-delay phenomenon is considered,which is an inherent characteristic in the implementation of neural network and it will lead to the oscillation of the system.Secondly,it is well known that the values of capacitance and resistance may be affected by unexpected random changes in the physical environment.In this case,the system parameters may fluctuate unnecessarily in practical engineering.Therefore,a group of random variables is proposed to represent the random fluctuations of system parameters.Thirdly,in the problem of state estimation,in order to avoid unnecessary waste of network resources,event triggering mechanisms,including static event triggering mechanism(SETM)and dynamic event triggering mechanism(DETM),are proposed based on the fact that the current information is transmitted only when certain triggering conditions are met.Finally,the measurement output is sometimes subject to abnormal interference(possibly due to unpredictable environmental changes or network attacks from opponents,etc.),which lead to abnormal measurement values.And if these outliers are directly incorporated into the innovation of the estimator design,the performance of estimator will be deteriorated.A certain confidence-dependent saturation function is introduced to weaken the effect of measurement outliers.In this thesis,some existing research results are extended to neural network system model,by using Lyapunov functional method and linear matrix inequality(LMI)technique to design state estimators for neural network systems based on event triggering mechanisms.Finally,MATLAB simulation examples are used to verify the effectiveness of the designed estimator.