【摘要】 随着第五代通信网络技术(5G)的发展,智慧城市中物联网(Internet of Things,IoT)的应用规模和多样性呈现出爆炸式增长.海量的智能传感设备组网给高动态的物联网通信服务质量带来了巨大的威胁.部分关键设备节点的失效以及网络攻击易引发物联网的链锁崩塌效应,影响网络应用的服务质量.因此,如何优化大规模物联网拓扑的鲁棒能力成为当下的研究挑战.目前,针对物联网拓扑结构的优化问题,研究者们提出了启发式算法、智能学习机制和多目标优化策略等创新方法提高物联网拓扑结构的鲁棒能力.但是,这些方法需牺牲巨大的计算资源来获得不成比例的鲁棒性能增益,网络规模越大,该现象越明显.为了解决这个问题并平衡计算开销和提升鲁棒性能,本文提出了一种基于网络模体(Motif)的轻量级物联网拓扑优化策略LITOS.首先利用物联网拓扑结构的社区属性,设计一种基于网络模体的异步社区发现算法,将大规模复杂拓扑结构分解为轻量级局部网络拓扑.然后,基于CPU多核心的计算资源,设计深度强化学习机制,异步优化轻量级物联网局部拓扑结构,从而降低网络整体优化运行时间,提高拓扑结构鲁棒能力.在实验方面,与其他先进的优化算法相比,该策略在运行时间方面降低了1～2个数量级,在鲁棒性提升方面,与最优算法相差大约10%.更多还原

【Abstract】 With the development of fifth-generation telecommunication technology(5G) and smart hardware devices,the scale and diversity of Internet of Things(IoT) applications are greatly expanding,like smart buildings,smart agriculture,smart homes,intelligent transportation,and so on.Numerous sensing devices can be networked to provide better network services for all aspects of life.However,the networking of massive intelligent sensing devices has brought a great threat to the quality of service(QoS) of the IoT,where the threat includes communication interrupts,data traffic congestion,and other unexpected factors.The robustness and performance of the IoT topology directly influence the network applications’ services and thus are associated with network reliability and resilience.The failure of key device nodes and network cyber-attacks can lead to the collapse of the IoT topology,which affects the QoS of the whole network system.Then,the system will finally fail without the ability to communicate with other network systems,where the large-scale IoT applications are broken.Therefore,how to optimize the robustness of large-scale IoT topology is a challenge to maintain maximum communication ability even if some device nodes fail,which draws researchers’ attention.Nowadays,methods such as heuristic algorithms and learning mechanisms are proposed to enhance the reliability of IoT topology.These methods can efficiently improve the robustness of IoT applications against cyberattacks and prolong the network lifetime even if part of the topology fails.However,these methods sacrifice huge computing resources to get disproportionately robust performance gains,and the larger the IoT topology scale,the more obvious this phenomenon is.Indeed,in the real world,IoT applications cannot spend so much time executing to produce a suboptimal result.We need a fast robustness optimization method for large-scale IoT applications.To address the problem,in this paper,we propose a Lightweight Topology Optimization Strategy for the IoT based on network motif(LITOS),where network motifs are significant repetition patterns widely spread in the network topology.Through the topology analysis,we found that the IoT topology has community characteristics,and we designed an asynchronous community detection algorithm based on a network motif to decompose the large-scale complex IoT topology into lightweight local network topologies.The devices with multiple roles in the communities are grouped together as a "super community".Then,utilizing CPU multi-core computing resources,we present a deep deterministic reinforcement learning mechanism(DDRL) to asynchronously optimize each local lightweight IoT topology,which can reduce the overall optimization time and improve the robustness of the network topology.Furthermore,we developed a novel robustness metric based on network motifs to measure dynamic changes in IoT topology,where network motifs can reveal hidden functional mechanisms and provide researchers with new perspectives on network topology.The new robustness metric has a better effect of guiding the topology towards more reliability.In terms of experimental results,compared with other state-of-the-art optimization algorithms,the running time of the LITOS is1-2 orders of magnitude lower than that of the ROCKS and ROSE algorithms and is about 10%lower than these algorithms in robustness improvement,which can greatly improve the optimizing efficiency for large-scale IoT topologies.更多还原