Research On State Estimation Of Neural Network System Based On Unreliable Information

As an active interdisciplinary subject,neural networks have attracted wide attention in the fields of system identification,pattern recognition,and intelligent control.After nearly a decade of development,academic research on neural networks has made remarkable progress.They are widely used in the fields of nonlinear system analysis and optimization,industrial applications,etc.,and have produced enormous practical value and benefits.However,with the rapid development of network technology,electronic technology,cloud computing technology and the urgent need to deal with big data,the combination of neural networks and intelligent network technology has brought breakthrough development and power to the research of neural networks,also caused new challenges and problems to the analysis and research work of neural networks in a networked environment.Therefore,this paper focuses on the problems of time delay,packet dropouts,stochastic nonlinearity in the neural networks model under the networked environment.The unreliable transmission due to the energy limitation of the sensor in harsh environments,and the problem of unreliable transmission of information,conducted and the following research work:1.For the discrete-time neural networks with distributed time-delay and unreliable measurements.When the measurement data is transmitted through multiple transmission channels,it is assumed that each channel is independent of each other and has different transmission packet dropout probabilities,and Markov chains are introduced to model packet dropouts of these channels.In order to simplify the calculation,the Markov chains of each channel are transformed into an augmented one,and then the multi-channel packet dropout situation a transformed into a load state dependent packet dropout model.Then the state estimator of network transmission state dependence is designed,and the state,the augmented form of the estimate and the output error are obtained.Furthermore,the channel-state-dependent estimator is designed,which effectively trade off between the number and the performance of the estimator.Sufficient conditions are established to guarantee that the augmented system is stochastically stable and satisfies the strict(Q,S,R)-?-dissipativity.Finally,an example is applied to illustrate the proposed estimator effectively balances the relationship between the number of estimators and the performance of the estimator2.The remote state estimation problem of neural network with energy acquisition sensor is studied.Considering the process of wireless transmission,there are sensor energy limited and packet dropouts caused by limit energy.Then the neural network state estimation strategy under unreliable transmission is designed.That is to say,the energy harvesting process of the sensor collector is periodic.According to the periodic characteristics of energy harvesting and the relationship between energy and packet dropouts rate,an energy level dependent packet dropout model is proposed.Periodic estimators are proposed to estimate the states of the system for the purpose of target tracking.Furthermore,the finite-horizon H_?performance is also guaranteed through the designed estimators.Finally,an example is used to clarify the effectiveness of the target tracking.Finally,on the base of summarizing the research content of this paper,the other problems existing in the neural networks model in the networked environment and engineering application are summarized and forecasted.