【作者】 孙扬；

【导师】 陶小峰；

【作者基本信息】 北京邮电大学 ， 电子与通信工程（专业学位）， 2022， 硕士

【摘要】 为满足移动设备日益增长的服务响应需求,边缘计算(Multi-access Edge Computing,MEC)作为一种新兴的计算模式被引 入以提升实时应用程序的用户体验。无线卸载是实现边缘计算的重要技术,由于无线信道的开放性,无线卸载面临无线窃听威胁,对用户数据安全造成严重威胁;同时,由于边缘节点的雾状地理分布特性,无线卸载方式极易暴露用户的位置隐私和使用模式隐私。这些问题极大限制了边缘计算的实用价值,因此,面向无线卸载的安全和隐私保护技术研究是近年来的研究热点之一。本文兼顾边缘计算无线卸载中的无线窃听和隐私泄露问题,从强化学习、遗传算法和凸优化等方面研究相应的解决方法。分别阐述如下:(1)基于强化学习的隐私保护和安全传输联合优化方法研究首先,针对边缘计算中本地参与任务处理的单用户单节点卸载场景,提出一种基于强化学习Q-Learning的隐私安全保护算法,可在保障隐私和传输安全的前提下,实现对MEC系统卸载平均能耗的优化。其次,基于马尔可夫决策过程建模系统的任务到达量以及无线信道的增益,描述卸载系统的隐私衡量值、假任务权重等关键参数对卸载性能的影响。最后,通过仿真分析发现,与随机卸载策略相比,所提方案具有更低的能耗表现,随着假任务权重的提升,隐私等级随之提升,所提方案的能耗和卸载率始终低于随机卸载策略。(2)基于任务分割与消息认证的安全隐私保护研究首先,针对边缘计算中因任务类型受到无限窃听者和恶意节点窃取而导致使用户模式隐私泄露的问题,提出一种基于任务分割和消息认证的隐私安全保护卸载策略。其次,基于隐私信源的信息熵,衡量卸载任务的隐私等级,采用遗传算法完成对卸载任务分割方案的选择,针对未知信道信息的窃听者,采用瑞利分布模拟其信道以确定既定中断概率下的安全传输速率。最后,通过仿真分析发现,与随机卸载策略相比,所提方案具有更低的平均相对熵,与回溯遍历卸载算法的时间复杂度O(2^N)相比,所提方案时延性能为O(N)。(3)面向异构网络合作并行卸载的安全隐私保护方法研究首先,针对边缘计算网络中多用户多节点卸载效率低下且隐私与安全难以保障的问题,考虑充分利用空闲节点的算力,以降低任务处理时延,提升用户使用体验,提出移动设备向多节点并行卸载的方案。其次,针对多用户多节点网络复杂的隐私和安全保护问题,本文以各移动设备分割后的任务信息熵作为约束,根据未知具体信道信息的窃听者信道特征控制卸载速率,将问题归结为含约束的非凸问题,并采用连续凸逼近的方法进行迭代求解,为大规模异构边缘计算网络提供了卸载方案。最后,通过仿真分析发现,与非合作方案对比,所提方案的并行卸载计算模式兼顾了隐私和安全的保护,同时具备更好的能耗和时延表现,且在节点与设备数量比值越高的场景下该优势越明显。更多还原

【Abstract】 Multi-access Edge Computing(MEC)is introduced as a new computing paradigm to improve information transmission capability in order to meet the increasing computing performance requirements of mobile devices.Wireless offloading is an important technology to realize edge computing.Due to the openness of wireless channel,wireless offloading faces the threat of wireless eavesdropping which poses a serious threat to data security.On the other hand,due to the vaporous geographical distribution characteristics of edge nodes,malicious nodes can collude with each other and estimate location and pattern of users according to the proportion of offloaded tasks to different nodes.These problems greatly limit the practical value of MEC.Therefore,security and privacy protection for wireless offloading is one of the research hotspots in recent years.The thesis considers wireless eavesdropping and privacy leakage problems in MEC.Corresponding solutions including reinforcement learning,genetic algorithm and convex optimization are studied.They are described as follows:(1)Research on j ointly resource optimization of privacy protection and secure transmission based on reinforcement learningAiming at the offloading scenarios that can be offloaded or processed locally,a privacy security protection algorithm based on reinforcement learning Q-Learning is proposed,which can optimize the average energy consumption of MEC system offloading on the premise of guaranteeing privacy and transmission security.Based on the task arrival quantity and wireless channel gain of the Markov decision process modeling system,the influence of key parameters such as privacy measure value and weight of fake task on offloading performance is described.Simulation analysis shows that compared with random offloading strategy,the proposed scheme has lower energy consumption performance.With the increasement of fake task weight coefficient,the privacy level increases,and the proposed scheme’s energy consumption and offloading rate are always lower than that of random offloading strategy.(2)Research on methods of enhanced privacy protection based on task segmentation and message authenticationFirstly,in order to solve the problem of user mode privacy leakage caused by eavesdropper and malicious node theft of task types in edge computing,a privacy security offloading strategy based on task segmentation and message authentication is proposed.Secondly,the privacy level of the offloaded tasks is measured based on the information entropy of the privacy information source,and the segmentation scheme of the offloaded tasks is selected by using the genetic algorithm.For eavesdropping of unknown channel information,the Rayleigh distribution is used to simulate its channel to determine the safe transmission rate under the given interruption probability.Finally,through simulation analysis,it is found that compared with random offloading strategy,the proposed scheme has lower average relative entropy,and compared with the time complexity O(2^N)of backtracking traversal offloading algorithm,the time complexity of the proposed scheme is O(N).(3)Research on optimization methods of security and privacy protection resources for cooperative and parallel offloading of heterogeneous networksAiming at the scenario of multi-user and multi-node in edge computing network,a scheme of mobile devices offloading to multi-node parallelly is proposed to make full use of the computing power of idle nodes to speed up task processing delay and improve user experience.The task entropy of each mobile device is used as the constraint,and the offloading rate is controlled according to the channel characteristics of the eavesdropper with unknown channel information.The problem is reduced to a non-convex problem with constraints,and the continuous convex approximation method is used to solve the problem iteratively.Simulation analysis shows that compared with the non-cooperative scheme,the proposed parallelly offloading computing paradigm obviously has better energy consumption and delay performance,and the advantages are more obvious in the scenario with a higher ratio of nodes to devices.更多还原