【作者】 张冬梅；

【导师】 马华东；

【作者基本信息】 北京邮电大学 ， 计算机应用技术， 2007， 博士

【摘要】 随着传感器节点技术的不断提高以及传感器网络应用的日益普及，人们对传感器网络提供的应用的灵活性和适应性要求越来越高，能够自适应环境条件变化和应用需求变化的大规模传感器网络由于其良好的协作能力和对复杂任务的支持能力而成为研究的热点。但由于传感器网络节点本身资源和能力的限制以及外界环境的多变性和不可预知性，无法将全部应用一次性部署在传感器网络中。因此，如何在资源受限的传感器网络上提供灵活多变的应用任务是制约传感器网络发展的一个重要问题。本论文以能量效率和环境自适应为目标，围绕无线传感器网络应用重构问题进行了研究，侧重于应用重构模型的建立、应用代码传输模式和相应的路由算法与传输协议的设计以及动态重构决策策略的制定等几个方面内容。论文工作主要包括：(1)针对传感器网络应用重构的环境相关性特点，利用知识推理方法，设计了一个具有环境自适应能力的无线传感器网络应用重构模型(Environment Adaptive Application Reconfiguration，EAAR)，实现了在资源受限的传感器网络上提供灵活多变的应用的目标。以EAAR模型为基础，结合传感器网络的分布式特性，设计了传感器节点主动触发的应用重构操作过程。(2)为实现EAAR模型中能量高效的代码传输，建模分析给出了EAAR模型中推(PUSH)、拉(PULL)模式代码传输的能耗关系，并结合两种模式的特点提出了一种适用于簇结构传感器网络的混合代码传输模式(Cluster-based Hybrid Code Transmission，CHCT)。在该模式下，簇头节点采用拉模式从汇聚节点获取代码，簇内节点采用推模式进行代码传输。(3)基于混合代码传输模式CHCT设计了传感器网络分层路由策略：簇头节点采用组播树传输代码，簇内节点采用洪泛路由。基于该策略提出了最小直径组播树(Minimum Diameter Multicast Tree，MDMT)算法构造簇头节点组播树。针对节点不同代码可靠性需求，设计了混合差错恢复机制。上述路由策略和算法在保证代码传输可靠性的基础上，节省了能量消耗。(4)利用马尔可夫决策过程对EAAR模型的重构决策过程进行了建模，提出了一种规则推理与强化学习相结合的动态重构决策系统框架；以能量约束和环境自适应性作为学习目标，设计了基于Q-学习的重构决策算法(Q-Learning Reconfiguration Decision Making，QLRDM)来调整规则的状态转移概率，使传感器节点的重构决策能够自适应环境的变化。(5)为了验证本文研究成果的可行性，发现实现过程中的具体问题，我们基于EAAR重构模型以及代码传输和重构决策的研究成果设计了原型系统。提出了原型系统的层次结构设计方案，分析选择了合适的移动代码中间件并重点描述了重构决策和模块动态加载的实现流程。更多还原

【Abstract】 With the rapid development of sensor techniques, potential applications of sensor networks span a wide spectrum. The flexibility and adaptation of applications provided by sensor networks have been paid more and more attention in recent years. Due to the ability of supporting cooperation among complex tasks, the large scale sensor networks which can self-adapt the change of environment and application requirement is becoming a hot issue. However, due to individual sensor’s limited resource and varied/unpredictable environment, we cannot deploy all the applications onto sensor nodes at one time. Hence, how to provide flexible and varied application tasks in resource-limited sensor networks is one of the most important issues which limit the development of sensor networks. Aiming at energy efficiency and environment self-adaptation, this thesis studies some fundamental issues of application reconfiguration in sensor networks, such as application reconfiguration model, code transmission paradigm, routing algorithm, transmission protocol, and dynamic reconfiguration decision scheme. The main contributions of this thesis are as follows:(1) Considering environmental correlation of application reconfiguration in sensor networks, we propose an environmentally adaptive application reconfiguration (EAAR) model using knowledge-based reasoning, thus provide flexible and varied applications in the resource-limited sensor networks. Combined with the distributed feature of sensor networks, we design the execution process of application reconfiguration triggered by the sensor node actively.(2) To achieve energy-efficient code transmission in the EAAR model, we analyze and compare the energy consumption relation of code transmission with the PULL and PUSH paradigms, and present a cluster-based hybrid code transmission (CHCT) for the cluster-based sensor networks. In this hybrid paradigm, cluster heads acquire codes from sink with the PULL paradigm, and cluster members transfer codes with the PUSH paradigm.(3) Based on cluster-based hybrid code transmission (CHCT), we propose a hierachical routing scheme in sensor networks. Cluster heads transmit code by a multicast tree, and cluster members adopt the flooding method. Based on this scheme, we propose a minimum diameter multicast tree algorithm (MDMT) to construct the multicast tree for cluster heads. For the different reliability requirements of sensor nodes, we design a hybrid error recovery scheme. The above routing algorithm and error recovery protocol can conserve energy while guaranteeing the code transmission reliability.(4) For decision making of the environment self-adaptation, we model the decision making in the EAAR model using the Markov decision process. We propose a dynamic decision making framework which combines rule-based reasoning with reinforcement learning. Aiming at energy constraint and environmental self-adaptation, we design a novel Q-learning based reconfiguration decision making algorithm (QLRDM) to adjust the state transfer probability of rules, thus the decision making of sensor node can self-adapt to environmental changes.(5) To testify the feasibility and effectiveness of our methods and find out the practical problems, we implement a sensor network prototype based on our EAAR model, code transmission and reconfiguration decision making schemes. We also present a hierachical architecture for this prototype, and select a mobile code middleware. Moreover, we discuss the implementation of reconfiguration decision making and dynamic module loading.更多还原