Research On Several Remote State Estimation Problems In Wireless Sensor Networks

With the rapid development of sensor technology,wireless sensor network has been widely used in a large scope of infrastructures.Wireless sensor network,which integrates micro-electro-mechanical technology,sensor technology,wireless communi-cation technology and distributed information processing technology,has always been the focus of research.As one of the most typical applications of wireless sensor net-work,target state estimation plays a huge role in many fields such as national defense,environmental monitoring,intelligent transportation,health care and industrial automa-tion,etc.On the one hand,the target state estimation based on wireless sensor network has the advantages of strong robustness and high estimation accuracy,but it is also subject to strict energy and bandwidth constraints.Existing works mainly focus on the optimal scheduling problem of a single system process or a single sensor under energy or band-width constraint,but the research on the optimal scheduling strategy under multiple system processes and multiple constraints and the distributed state estimation of mul-tiple sensor nodes is seriously insufficient For example,the existing distributed state estimation algorithm only considers the case of traditional fixed topology and ignores the constraint of limited sensor energy.On the other hand,the wireless sensor network is usually deployed in the unmanned maintenance and uncontrollable environment,so it will face a variety of threats such as denial-of-service attack,deception attack and cause information loss,information tampering,etc.It is impossible for users to accept and deploy a sensor network that has not solved the security problems well.Therefore,in the state estimation of wireless sensor network,it is necessary to fully consider the security problems that the wireless sensor network may face and integrate the security mechanism into the system design.Existing works about state estimation under denial-of-service attack are mainly focused on the optimality problem from single a point of view.Designing defensive or offensive strategy that takes into account the effects of both the sensor's and the attacker 's actions on the system is complex,and research on this is scarce.Therefore,this thesis is aimed at these deficiencies and carries out research which mainly includes two aspects:(1)the optimal scheduling problem and distributed state estimation of wireless sensor network where the communication environment is faced with multiple constraints such as bandwidth and energy;(2)The performance analysis of wireless sensor network under attack and the design of the corresponding attack and defense strategy.The specific works and innovations of this thesis are as follows:1)In Chapter 2,state estimation of wireless sensor network with limited resources is investigated.We design an optimal scheduling scheme for second-order Gaussian-Markov system with limited resources.The two cases that the sensor has sufficient calculation capability and the sensor has limited calculation capabil-ity are considered.Also,a necessary condition for the optimal sensor scheduling scheme is given.Based on this necessary condition,an explicit periodic opti-mal scheduling method is proposed under the constraints of transmission power and channel bandwidth,and it is strictly proved that the estimation error of this method in the remote estimation center is minimal.2)In Chapter 3,distributed state estimation in wireless sensor network with limited resources is investigated.A novel distributed state estimation method based on randomized gossip algorithm is designed to solve the scenario of time-varying topology.The randomized gossip algorithm requires that the topology of sensor nodes is connected but not fixed.However,the difficulty brought by the random-ized gossip algorithm is that the convergence and performance analysis are based on the probabilistic sense.This thesis will build a new distributed estimation algorithm convergence analysis method and compare the performance with the existing distributed estimation algorithm.Under a sufficient condition,we show that the proposed algorithm is quite appealing as it achieves better mean square error performance theoretically than the noncooperative decentralized Kalman filtering algorithm.We also prove th at our proposed algorithm possesses better performance for a particular class of setups such as scalar systems.3)In Chapter 4,optimal energy control for wireless sensor network under denial-of-service attack is investigated.Considering that the actions of attacker and sensor are interactive in a real scenario,the general-sum stochastic game model is intro-duced to describe this conflict characteristics.In addition,different from most of the existing works which are based on the stationary channel environment,this thesis adopts the finite state Markov chain model to investigate the optimal energy control problem in time-varying channels.An optimal strategy in Nash equilib-rium point is derived through reinforcement learning algorithm.At the same time,under a sufficient condition,the optimal stationary strategy with monotonic struc-ture is constructed.Finally,this thesis adopts the framework of Bayesian game to analyze the cases where some channel state information is unknown and obtains a Nash equilibrium strategy based on its own channel information.4)In Chapter 5,performance analysis of wireless sensor network under deception attacks is investigated.The concepts of strictly stealthy deception attack and e-stealthy deception attack are proposed for the first time in this thesis.For the strictly stealthy deception attack,we present a necessary and sufficient condi-tion for the existence of this strictly stealthy deception attack.A design method for the attacker to achieve the strictly stealthy deception attack is also provided.Considering that the strictly stealthy attack strategy is associated with the un-detectable points of difference equations,we propose an algorithm to find such points.The corresponding defense strategy can be designed by avoiding the ini-tial state falling into this undetectable set.While for ?-stealthy attack,a necessary condition is provided for its existence.Furthermore,we prove that the e-stealthy attack for the difference system does not exist as system matrix does not have an unstable eigenvalue.