Dissipative State Estimation Of Discrete-Time Stochastic Memristive Neural Networks Under Cyber Attacks

With the deep research of biological neural network,artificial neural network arises at the historic moment,which opens up a new research direction for nonlinear complex system modeling and analysis.In addition,it is noted that artificial neural network,as the cornerstone of artificial intelligence technology,has once again attracted the attention of scholars and the industry.However,as artificial intelligence technology has entered a new era that is more influenced by"brain-like technology",artificial intelligence presents more obvious characteristics of platform,chip and hardware.CMOS,which is the base of traditional artificial neural networks,has been difficult to meet the needs of processing massive data because of its physical defects.At this time,the emergence of memristor,as a substitute for CMOS,brings new hope to solve the problem.Memristor,as a kind of storage device,has the ability of information storage and processing similar to human brain synapses.The neural networks based on memristors have the advantages of higher integration and lower power consumption,which breaks the bottleneck of traditional neural network research and attracts more and more attention.The dynamic research of memristive neural networks have become a research hotspot.In engineering practice,memristive neural networks are inevitably affected by the complexity of internal system and external noise,which makes it difficult to know the neuron information,thus affecting the system can not achieve the expected goal.In view of this,how to obtain effective neuron state has become particularly important,and has become a hot issue concerned by scholars.In this paper,the dissipative performance and state estimation of discrete delayed memristive neural networks are studied.The specific works are as follows:(1)The state estimation of discrete-time delayed memristive neural networks subject to deception attacks is studied.When the system is attacked,the system cannot receive the correct real-time measurement information,which leads to the decrease of network stability and even the paralysis of the whole network.For deception attacks,a set of Bernoulli distribution white sequences with known conditional probability is introduced to describe the randomness of attacks.In addition,considering the network constraints,in order to further improve the bandwidth utilization,a dynamic event-triggered mechanism is introduced to reduce unnecessary resource consumption in the communication channel.The strict dissipative state estimation problem of memristive neural networks affected by deception attacks is studied.A co-design scheme of estimator parameters and the event-triggered matrix is proposed.Sufficient conditions are obtained to make the estimation error dynamically stochastic stable and strictly dissipative.Finally,the effectiveness of the proposed state estimation scheme is verified by numerical simulation based on linear matrix inequality.(2)The state estimation of discrete-time delayed memristive neural networks under denial-of-service attacks is studied.A set of Bernoulli distribution white sequences with known conditional probability is used to reflect the randomness of denial-of-service attacks.In this chapter,the extended dissipativity,a more general performance index,is considered,and the stability and extended dissipativity of discrete-time delayed memristive neural networks are analyzed.A new estimator is designed so that the error system is extended dissipative.Finally,an example is presented to prove the superiority of the proposed scheme.(3)The H_∞state estimation problem of discrete-time delayed memristive neural networks with denial-of-service attacks is studied.The previous research results are extended,and the denial-of-service attacks are described by attack frequency and average dwell time.According to the switched system theory,the system is considered as a switched system in which two states of active attack and silent attack are switched each other.Then,based on Lyapunov stability theory,sufficient conditions to guarantee the exponentially mean-square stability and H_∞performance of the error system are obtained.Finally,the effectiveness of the proposed method is demonstrated by numerical simulation results.