H_? State Estimation Of Discrete Time-varying Recurrent Neural Networks With Variance Constraints

As it is well known,the artificial neural network(ANN),which is simulated based on the complex structure and function of cells in the human brain,is a class of information processing systems.Due to its advantages of strong associative ability,fault-tolerant ability and adaptability,the ANN can be applied in many practical fields such as pattern recognition,signal and image processing.However,during the process of practical application,the information of neuron is often not completely measurable.In order to know the internal operation states of recursive neural network and master the reliable state information of network,it is of great theoretical and practical significance to estimate the internal states of neurons effectively.Therefore,based on Lyapunov theory and stochastic analysis method,the H_? state estimation method with variance constraint is proposed for several classes of time-varying recursive neural network systems in this thesis.The main contents are listed as follows:Firstly,the H_? state estimation problem is investigated for a class of discrete-time varying recursive neural networks with randomly occurring nonlinearities.Here,a random variable obeying the Bernoulli distribution is employed to characterize the randomly occurring nonlinearities.A novel time-varying state estimator is designed.Later,by utilizing the stochastic analysis method,sufficient conditions are given for the augmented system to satisfy both the upper bound of error variance and the specified H_? performance requirements.Then,a new time-varying state estimation method is proposed.Finally,a simulation example is given to verify the effectiveness of the proposed H_? state estimation scheme with variance constraint.Secondly,the H_? state estimation problem is studied for a class of discrete time-varying recursive neural networks with randomly occurring nonlinearities and missing measurements.Here,some random variables obeying the Bernoulli distribution are employed to describe the randomly occurring nonlinearities and missing measurements.Based on the probability information of missing measurements and the stochastic analysis method,some sufficient conditions are obtained for the augmented system to satisfy both the upper bound of the error variance and the specified H_? requirement.Finally,a simulation example is given to verify the feasibility of the proposed algorithm.Thirdly,the resilient H_? state estimation problem is discussed for a class of discrete time-varying recursive neural networks with random saturation.Here,a random variable subject to the Bernoulli distribution is employed to describe the random saturation,for the random saturation,a new resilient state estimator is designed,and some criteria are given to guarantee that the augmented system satisfies both the upper bound of the estimated error variance and the specified H_? performance requirements.Similarly,the effectiveness of the proposed H_? state estimation method is tested by numerical simulation example.Fourthly,the resilient state estimation problem is discussed for a class of discrete time-varying recursive neural networks with event-triggering and fading measurements.Some random variables obeying the Bernoulli distribution are employed to describe the fading measurements,where each sensor has its own loss probability is characterized.Based on the acquired information,a novel event-triggered H_? state estimator is designed and a sufficient criterion is obtained to ensure that the augmented system satisfies both the upper bound of error variance and the specified H_? performance requirements.Finally,the superiority of the new event-triggered H_? state estimation scheme is verified by the numerical simulation.