Research On Security And Privacy Of Consensus Algorithms In WANETs

With the rapid advancement of sensing,computing and wireless communication technologies,a new networking paradigm called Wireless Ad Hoc Networks(WANETs)come into being.Thanks to its flexibility,mobility,adaptivity and scalability,WANETs have triggered a revolution in the field of information perception and fusion.With its typical implementations in wireless sensor networks,vehicular ad hoc networks and online social networks,WANETs is changing the way of information propagation and integration tremendously and it has gained great attentions from global researchers.By modeling a node in WANETs as an intelligent agent capable of sensing,computation and communication,nodes can reach global consensus via local interactions.Multi-agent consensus,especially average consensus,is the supporting algorithm for a lot of WANETS-based applications.In the last two decades,researchers have conducted a comprehensive and in-depth theoretical study of multi-agent consensus algorithms.However,there is still a lack of research concerning network security and data privacy.One aspect is that absence of evaluating node importance in average consensus makes it difficult for system designers to locate network bottleneck.Meanwhile exist-ing average consensus under attack has no guarantee of the convergence value.The other aspect is that the data privacy in consensus is not fully preserved,especially when each node holds separately a giant amount of data.In this thesis,based on the latest results,node importance in average con-sensus is evaluated and robust median consensus considering data privacy are studied.The main work and contributions can be summarized as follows.1.A brief introduction of multi-agent consensus problems and related work in WANETs is pro-vided,especially the work concerning data privacy and network security.2.For the possible physical destruction attack on node in WANETs,node importance in average consensus is evaluated.By modeling node importance as the algebraic connectivity descent when removing designated node,three centralized node importance criteria are proposed with theoretical error bounds analysis.Furthermore,such centralized criteria are approximated to yield three distributed node importance evaluation algorithms.Distributed power iteration combined with initial vector adjustment and event-triggered scaling is utilized to realize the distributed node importance evaluation.3.A fast median consensus algorithm in ad hoc network is designed to improve the robustness to false data injection attack in average consensus.Based on finite-time average consensus algorithm,a binary search based median consensus called FAMB is firstly designed.FAMB achieves accurate but asymptotic median consensus in a geometrically fast way.Based on FAMB and the unique ID assumption,an even faster finite-time(2nd)accurate median con-sensus called FAMC is proposed.FAMC requires less local computation and storage con-sumption(compared with Flooding)and faster convergence(compared with distributed op-timization).4.For a WANET where each node holds heterogeneous data and the data scale in each node is private,an efficient privacy-preserving median consensus algorithm DIME is proposed.Specifically,median consensus is modeled as a distributed l1 norm minimization problem.Combined average consensus with local gradient descent,a weighted average of local histor-ical states approaches the median asymptotically.The step size principle is given with strict convergence proof and error analysis.5.For the fusion of large-scale geo-distributed sensitive data,a privacy-preserving k-selection algorithm is proposed,with DPAM tailored to median selection as a first step.Since median is at most one standard deviation away from median,distributed server nodes share with each other noisy local statistics to decrease the standard deviation while preserving original median,yielding an accurate median approximation via mean.Furthermore,?-differential privacy is used to devise noise mechanism to protect local statistical privacy.DPAM has the theoretical lowest computation complexity and very low communication complexity.In the end,the thesis is concluded and some further research directions are discussed.