Study On Synchronization Of Memristive Neural Networks With Time Delays

The memristors are considered to be a kind of ideal device for constructing synapses due to their advantages such as memory property and low power consumption.The neural network whose synapses constructed by memristors is called memristive neural network(MNN),which is an important research topic of neuromorphic computing at present.In addition,MNN is also a special kind of complex network,which has the dynamic peculiarities of complex network,among which synchronization is a common one.By designing suitable controllers,MNN can achieve different types of synchronizations.MNN and its synchronization control have significant applications in the fields of coordinated control of intelligent agents,pattern recognition and brain function cognition.In addition,due to the unique hysteresis loop of memristor,MNN usually has rich dynamic peculiarities and can generate some new types of complex chaotic attractors.In light of its dynamic peculiarities,the synchronization of MNN is also applied in the fields of image encryption communication and pseudo-random number generation.Aiming at structure of current synchronization model,network security in practical encryption application and timeliness of quasi-synchronization,this paper studies weighted output synchronization,the synchronization under two types of cyber-attacks and finite-time quasi-synchronization for time-delay recursive MNN,respectively.The following are the specific works and innovations of the paper.(1)Weighted output synchronization model is proposed for MNN,which includes two works.In the first work,because current synchronization models only focus on the behavior of single neuron,weighted output synchronization of MNN is investigated when considering the importance of weighted combination behavior of neurons in neural networks.State feedback control and aperiodic intermittent control are utilized to realize weighted output complete synchronization and quasi-synchronization,respectively,and some switching parameters are introduced to improve the flexibility and the anti-interference ability of control system.Finally,simulations are employed to verify the validity of theoretical analysis.Considering that neural networks are a class of clustered system,the second work further investigates cluster-based weighted output synchronization.Firstly,since there is no related research,a network model consisted of multiple clusters is designed.Then,adaptive control and state feedback control are used to realize cluster-based weighted output synchronization of MNN.Finally,simulations are employed to verify the validity of theoretical analysis.Compared with previous synchronization models,weighted output synchronization pays more attention to cooperative behavior of neurons,which provides a novel standpoint and reference for the researches on synchronization of neural networks.(2)Because the existing literatures ignore network security in practical encryption applications,quasi-synchronization of stochastic MNN under deception attacks is studied,where hybrid impulse control is used to realize system synchronization.Firstly,as previous differential inequality cannot effectively handle quasi-synchronization problem of impulsive system,an inhomogeneous differential inequality is proved,which is general and can be also used to study other dynamical behaviors of impulsive systems.Then,by using this inequality and Lyapunov stability theory,quasi-synchronization criterion and error bound under the attack are obtained.In addition,without the attack,complete synchronization is investigated.Based on the obtained results,we analyze the effect of deception attacks and explore how to mitigate this effect.Finally,simulations are employed to verify the validity of theoretical analysis.(3)Synchronization of MNN under Do S attacks(Denial-of-service attacks)is investigated,and then the synchronization results are applied to image encryption communication.By utilizing observer-based control strategies and Lyapunov stability theory,some sufficient conditions to ensure the synchronization under Do S attacks are obtained.In addition,the influence of actuator saturation is considered in the controller design,so that the controller output is limited,which is more in line with actual output.Finally,an encryption scheme for color image is designed based on the obtained synchronization results.Compared with the previous schemes,theoretical and experimental results show that our scheme has more reliable encryption performances under Do S attack and actuator saturation.(4)In light of the limitations of exponential and asymptotic quasisynchronization models studied in the current literature,and considering that practical application has a time-efficient requirement,finite-time quasisynchronization of MNN is studied.A quantized event-triggered intermittent control scheme is designed to guarantee that the synchronization error converges to the specified error bound within finite time.By using Lyapunov stability theory and proofing two new differential inequalities,we obtain synchronization criteria,error bound and synchronization time.Compared with exponential and asymptotic quasi-synchronization models,finite-time quasisynchronization model can ensure that quasi-synchronization can be achieved within finite time and thus it is more practical.