State Estimation For Discrete Delayed Recurrent Neural Networks With Missing Measurements And Signal Quantization

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.The ANN has received close attention from scholars because of its ability of solving the practical problems effectively in many fields such as signal processing,pattern recognition and so on.The ANN can be implemented by the circuit system.With the aid of the memristive,the memristive recurrent neural network has also received wide attention.On the other hand,because of the influences of external noise disturbances and equipment itself in a real environment,the information of the neuron is not completely measurable.Therefore,it is of important research significance and application values on how to estimate the internal running state of the system based on the available measurement information.In this thesis,we aim to consider several classes of discrete recurrent neural networks with missing measurements and stochastic signal quantization,and develop some new robust state estimation schemes via the linear matrix inequality technique.The main contents are summarized as follows:Firstly,the non-fragile state estimation problem is studied for a class of discrete recurrent neural networks(DRNNs)subject to mixed time-delays and missing measurements.The mixed time-delays include the randomly occurring time-delay and distributed sensor delays.First of all,a state estimator is designed for the system based on the available measurement information.Subsequently,an appropriate Lyapunov-Krasovskii functional is constructed.By combining the relevant inequality techniques and Lyapunov stability theory,a sufficient condition is established to guarantee that the augmented system is exponentially stable in the mean square and the explicit expression of the desired estimator gain is proposed.Finally,a numerical example is used to verify the validity of the presented state estimation method.Secondly,the finite-time H? bounded state estimation problem is investigated for a class of discrete memristive recurrent neural networks(DMRNNs)with randomly occurring time delay and missing measurements.Due to the state dependence of the system parameters of the MRNNs,the system parameters are first processed.Then,a new state estimator is designed in a view of the probability of stochastic delay and missing measurements.A sufficient condition is given to ensure the finite-time H? boundedness of the augmented system and the explicit expression of the state estimator gain is obtained based on the Lyapunov stability theory.In addition,the feasibility of the proposed state estimation strategy is verified by a numerical simulation example.Thirdly,a resilient approach to the finite-time bounded estimation is provided for a class of DMRNNs subject to stochastic quantized measurements and multiplicative noise.Here,the multiple additive noises are employed to characterize the estimator gain variations.Based on the available measurement information,a non-fragile state estimator is constructed.And then,a sufficient condition is derived to ensure that the recurrent neural network is finite-time bounded based on the Lyapunov stability theory and stochastic analysis theory.Moreover,the explicit form of the non-fragile estimation gain is presented based on the solutions to certain matrix inequalities.Similarly,the feasibility and validity of the developed estimation method are verified by numerical simulation results.