The Characteristic Of Neuronal Network: An Identification And Control Theory Prospective

Brain is a complex large scale network composed of millions of neuronalnetworks with different kinds of topology. The changes of parameters or topology inneuronal networks may lead to abnormal actions of neural system, which might be themechanism of neural diseases. Neural electrical stimulation methods such as DBS andTDCS are effective therapies to cure mental diseases by stimulating the brain to adjustthe brain activities back to normal condition. During the stimulation process of curingmental diseases, it will be helpful to achieve better effect if we can make theparameters in neural system, the structure of neuronal networks and the input-outputrelationship clear. So in this dissertation, we study the system composed of neuronmodels and network models in the view of control theory to analyze the effect ofparameters and topology of neuronal networks. Moreover, some methods based onsystem identification and control theory are proposed to solve the problems ofparameter and state estimation, model rebuilding and dynamical adjustment in asingle neuron and neuronal networks.First the effect of neuron parameters and network topology on thesynchronization of neuronal networks is studied in this dissertation. By analyzing thesimulation results of34three-neuron networks with different kinds of topology, theconclusion that the self-coupling in a neuronal network will prevent neurons in thenetwork from becoming synchronized is drawn and proved.During the control process of neural system, it is nearly impossible to detect allparameters and topology of neuronal networks. According to this problem we proposea "bridging network observer" method. With the detected error between the actionsand states of the healthy network and the abnormal network, the "bridging networkobserver" can successfully estimate the values of unknown parameters and thetopology of the healthy network. With these idenfied parameters and topology, we candesign corresponding methods to adjust the abnormal network back to normalcondition.Challenges such as long sampling period, undetectable inner structure of neuronsor high disturbance exist during the process of neuron modeling. To solve thisproblem, some methods about parameters and states estimation are given in Chapter5.First we combine the mid-value estimation method with adaptive estimation methodto solve the problem that the sampling period is too long to satisfy the accuracy of estimation. Then fuzzy estimator and ANN estimator are used to rebuild dynamicmodel for the actions of neurons. Finally, an observer based on the idea of ANN isproposed to estimate and eliminate the disturbance in neurons.Moreover, at the end of this dissertation, we focus on solving the problem ofdesigning neural electrical stimulation signals. First a closed-loop controller based onadaptive parameter estimation is used to control H-H model exposed in the electricfield to explain the mechanism of electrical stimulation. Then two closed-loopmethods which can be also used without knowing the math model of neuronpopulation are proposed. One is the closed-loop robust iterative learning algorithmwhich can make the target neuron population act periodically as we desired; the otheris recursive least-square method to rebuild the input-output relationship in singleneuron or neuron population and design the control signals. With these two methods,the stimulation signals can be designed without knowing the model of the targetneuron population.The results in this dissertation can be considered as the theoretical basis ofmental disease diagnosis and therapy (such as DBS and TDCS) in the future.