Performance Analysis And Control Design Of Several Neural Networks With Delay In Network Environment

In this era of rapid development of information network,as an indispensable part of modern artificial intelligence technology,neural networks have been extensively studied and successfully applied to a variety of science and engineering fields,including infor-mation field,medical field,control field,transportation field,etc.It is worth noting that these practical applications of neural networks depend on their dynamic performance to a great extent.Nevertheless,due to the limitation of signal transmission speed between neu-rons and some external interference factors,time delay is common in the implementation and application of neural networks,which may lead to unexpected dynamic behaviors,such as oscillation,poor performance,or even instability.In addition,under the network environment,due to equipment aging,strong electromagnetic interference,network con-gestion and external environmental interference,sensor or actuator failure is inevitable when the neural networks to perform the special tasks,which may lead to system instabil-ity or even catastrophic accident.Therefore,how to design an effective control strategy to maintain the neural networks stable is a hot research topic,which has aroused more and more research interest from various fields,especially system science and control field.Based on the above analysis,it is very necessary to study the performance analysis and control design problems of delayed neural networks from both the theoretical significance and practical application.At present,although some preliminary achievements have been made in this field,but there is still a lack of systematic theoretical research in some as-pects.On the basis of the existing work,this dissertation further improves the theoretical research system in this respect.In this dissertation,the performance analysis and control design problems of sev-eral kinds of delayed neural networks with random faults in the network environment are studied by adopting several different control strategies.The specific research problems in-cludes:1)Semi-Markov jump actuator faults-based nonfragile sampled-data synchroniza-tion control problem of fuzzy delayed neural networks;2)Probabilistic sensor faults-based event-triggered control problem of continuous-time semi-Markov jump neural networks;3)Probabilistic sensor faults-based event-triggered guaranteed cost control problem of discrete-time semi-Markov neural networks;4)Probabilistic sensor and actuators-based finite-time H? control problem of discrete-time fuzzy neural networks.The main research outcomes of this dissertation are as follows:1.Applying the scheme of nonfragile sampled-data control with communication de-lay,the synchronization problem of T-S fuzzy neural networks with semi-Markov jump actuator faults is discussed.To characteristic the fault behavior of actuator accurately,an input model with semi-Markov actuator fault is presented,which includes the linear term and randomly occurring bias term.Unlike most sampled-data control strategies,a con-stant transmission delay signal is introduced in the feedback loop and the sampling period is assumed to vary within an interval.By selecting an appropriate Lyapunov function and combining with the N-order integral inequality technique,the H? asymptotic stability cri-terion of the synchronization error system of fuzzy neural networks is developed.Then,by means of the method of variable substitution and the assumption of the upper and lower bounds of the transition rates,the conditions for the mode-dependent non-fragile fuzzy control gain matrix parameterization are exhibited.Ultimately,a numerical example is presented to substantiate the effectiveness of the proposed control method,which shows the feasibility of the proposed method.2.Introducing the event-triggered communication mechanism,the stochastically mean square exponential stability problem of networked continuous-time semi-Markov neural networks is studied.Probabilistic sensor faults-based event-triggered mechanism is designed,and probabilistic actuator faults are considered simultaneously,and then a control strategy based on probabilistic sensor and actuator faults is developed.By exploit-ing Lyapunov function method,free-weighting matrix and stochastic analysis technique,the criterion of the mean square exponential stability for semi-Markov neural networks is formulated.Then,the controller design scheme is provided by utilizing linear matrix inequality(LMI)decoupling technique.Lastly,the validity and applicability of the devel-oped methodology are determined by a quadruple-tank process system.3.Considering the network communication delays,the guaranteed cost event trig-gered control problem of networked discrete-time semi-Markov neural networks is inves-tigated.By constructing a mode-dependent Lyapunov-Krasovskii functional,and apply-ing free-weighting matrix and convex combination technique,the conditions of stochastic stability for the closed-loop discrete-time semi-Markov neural networks are established.Then,the solution of the optimal upper bound of guaranteed cost control is transformed into a convex optimization problem,which constrained by a series of linear matrix in-equalities(LMIs),and the design scheme of guaranteed cost controller is given.Finally,the effectiveness and applicability of the theoretical results are verified by the discretized memristor-based neural networks.4.Adopting state feedback control strategy,the finite time H? control problem of networked discrete-time fuzzy neural networks is addressed.Considering the probabilis-tic sensor and actuator faults,as well as the communication delays of the pre-transmission data from sensor to controller and controller to actuator,a reliable delayed state feedback control scheme is developed.By employing Lyapunov functional method,free-weighting matrix method and inequality technique,the criterion of finite time H? boundedness for discrete-time fuzzy neural networks is established.Based upon the obtained conditions,by eliminating the nonlinear coupling terms in the matrix inequalities,a method for cal-culating the state feedback control gains is presented.Eventually,the effectiveness and applicability of the proposed method are verified by simulation with fuzzy genetic regu-latory networks(FGRNs)as the control object.