State Estimation And Quasi-synchronization Of Neural Networks With Information Restriction

In recent decades,the research of neural networks has been developed rapidly,and the neural networks technology has been improved unprecedentedly.These have laid a solid foundation for the application of neural networks in practical engineering.Neural networks are widely applied in many fields such as driverless,financial forecasting,image processing,and so on.There are many practical problems with the integration of networked systems into neural networks,such as transmission induced delay,quantization induced error,external interference induced uncertainties,channel transmission induced fading and so on.These problems not only affect the performance,but even the stability of the system.Synchronization phenomena are widespread in nature,and the result is probably good or bad.Studying the synchronization problem among multiple systems of neural networks can expand the application of good behaviors on the one hand,and avoid bad behaviors on the other hand.In practical engineering,parameter mismatch caused by system characteristics is a common phenomenon.Therefore,it is of great significance to study the quasi-synchronization problem of the system.This thesis studies the stability,state estimation and quasi-synchronization of neural networks in the networked environment.The main research contents of this thesis are as follows.(1)The state estimation of discrete neural networks for Markov jump weight matrix and transmission delay has been studied in chapter two.Markovian jump interval matrices have been introduced to model the uncertainty of the connections among neurons.The mode-dependent transmission delays have been introduced to describe an unideal communication channel.Sufficient conditions of stochastic stability and strict-(Q,S,R)dissipative performance have been derived for the augmented system.Then mode-dependent estimator gains have been designed.At last,an example has been employed to illustrate the obtained results.(2)The problem of the estimator design for the periodic neural networks with polytopic uncertain connection weight matrices has been studied in chapter three.The polytopic uncertainty has been used to model the uncertain weight matrices.Bernoulli processes have been employed to characterise the randomly occurred sensor nonlinearities.Lyapunov function which depends both on the polytopic vertices and the period has been constructed to improve the performance of the estimator.Sufficient conditions of the stochastic stability with H? performance for the augmented system have been established,and the corresponding gains of the estimator have been designed.Finally,an illustrative numerical example has been given.(3)The problem of state estimation for time-varying neural networks under sensor energy constaint has been studied in chapter four.The packet dropout conditions of main channels have been modeled by Bernoulli processes.To reduce the packet dropout rate and prolong lifttime of sensor,a low-energy transmission and retransmission of packet dropout occurring data transmission strategy has been introduced.Sufficient conditions have been obtained to ensure that the augmented system satisfies the l2-l? performance over finite-horizon,and the estimator gain design algorithm has been obtained.At last,a numerical example has been shown to verify the derived results.(4)The quasi-synchronization for master-slave neural networks with parameter mismatch has been studied in chapter five.The connection weight matrix of master neural networks is randomly disturbed and has been modeled by a norm-bounded uncertainty.Sufficient conditions of quasi-synchronization for the synchronization error system in the mean sense have been established,and the controller design algorithm has been obtained based on this sufficient condition.Finally,the proposed result has been illustrated by a numerical example.(5)The problem of quasi-synchronization for Markovian jump master-slave neural networks with time-varying delay has been studied in chapter six,where mismatch parameters and unreliable communication channels have been also considered.A set of stochastic variables with different expectations have been used to describe the fading phenomenon of parallel communication channels.An impulsive driven transmission strategy has been designed to reduce the communication load,and a corresponding impulsive controller thus has been designed.The synchronization error system has been obtained,and a convex quasi-synchronization condition has been established for the synchronization error system.A linear matrix inequality based iterative algorithm has been proposed to minimize the bound of the synchronization error system,and the controller gains have been calculated.A numerical example has been provided to illustrate the effectiveness of the developed result.