【作者】 牛建军；

【导师】 邓志东；

【作者基本信息】 清华大学 ， 计算机科学与技术， 2010， 博士

【摘要】 无线传感器网络本质上是一类资源受限的网络。通常情况下,无线传感器网络节点采用电池供电,有限的能量限制了网络生存期。作为一个嵌入式系统,节点的计算能力和存储能力都较小;节点间通信带宽也比较低。因此,在这些约束条件下,无线传感器网络调度方法对于WSN应用中网络性能的提升具有极其重要的意义。本文对无线传感器网络分布式调度方法进行了深入的研究,并针对无线传感器网络能量受限的特点,提出了四个分布式调度方法。论文的主要贡献和创新点包括:(1)提出了一类实用的、协作分布式的调度方法。该类方法包含两种调度方法:一步协作分布式调度方法和两步协作分布式调度方法。该类方法基于马尔可夫链,使节点通过相互协作学习其它节点的行为信息,将节点传输调度与休眠调度相结合,达到节省能量的目的。分析了该类方法的适应性、实用性,并从理论上证明了其收敛性。实验结果表明这两个调度方法能够有效地减少能量消耗。(2)提出了一种WSN分布式自学习调度方法。该方法在WSN分布式调度研究中引入再励学习的思想,通过对Q-学习方法扩展,提出了一种计算调度参数的近似方法,使节点可以得到具有连续值空间的发送调度参数和休眠调度参数,从而使节点实现传输调度和休眠调度。该方法在优化节点能耗的同时,还可以减少数据包时延。在MAC层上实现了该调度方法。仿真结果表明,该方法可以有效地节省能量,并减少了树型拓扑网络的数据包平均时延。(3)提出了一种WSN分布式进化自学习调度方法。该方法根据WSN中具有相同父节点的近邻节点之间工作特性具有一定相似性的特点,对这些节点的调度策略进行优化。并且针对无线传感器网络节点带宽有限的特点,提出了一种将粒子群优化算法与分布式自学习方法相结合的方法,使节点可以从其它节点学习调度的经验,加快了调度策略的学习速度。仿真结果表明,该方法比占空比10%的S-MAC更节省能量,而数据包时延与占空比60%的S-MAC协议相接近。与分布式自学习调度方法相比较,在仿真结束时,可以进一步节省能耗,并减少数据包平均时延。更多还原

【Abstract】 Wireless sensor network (WSN) is virtually a resource constrained network system. In most cases, WSN nodes are battery-powered, which restricts the lifetime of WSN. As an embedded system, WSN nodes are also characterized by restrained computation capacity and storage capacity. Moreover, communication bandwidth is much narrower among WSN nodes. Therefore, given these restrictions, it is significantly important to adopt scheduling approaches into the real applications of WSN in order to improve the network performance.This dissertation conducts an in-depth study of the distributed scheduling approach for wireless sensor network, and proposes four effective distributed scheduling approaches to address the issues of restrained energy. The main contents and contribution of this dissertation are given below:(1) A family of collaborative distributed scheduling approaches (CDSAs) is proposed to reduce the energy consumption of WSN. The family of CDSAs comprises of two scheduling approaches, i.e. one-step collaborative distributed scheduling approach (O-CDSA) and two-step collaborative distributed scheduling approach (T-CDSA). The family of CDSAs, based on the Markov chain, enables nodes to learn the behavior information of other nodes collaboratively and integrate sleep scheduling with transmission scheduling to reduce the energy consumption. In this dissertation, the adaptability and practicality features of the CDSAs are analyzed and the convergence feature is proved. The test results show that the two proposed approaches can effectively reduce nodes’energy consumption.(2) A distributed self-learning scheduling approach (SSA) is proposed in order to reduce energy consumption and latency for wireless sensor network. This approach employs reinforcement learning algorithm in scheduling approach of WSN and extends the Q-learning method. Based on a proposed approximate computational method of scheduling parameters, SSA enables nodes to learn continuous transmission parameters and sleep parameter through interacting with the WSN. Then, the WSN nodes can schedule its sleep and transmission through this approach. We implement the SSA in a MAC protocol. The simulation results show that the SSA can effectively reduce energy consumption, and can reduce the average latency of data packs in tree topology network.(3) A distributed evolutionary self-learning scheduling approach (ESSA) is proposed in order to reduce energy consumption and latency for wireless sensor network. Considering the similarities of working features among the nodes, which have the same parent node and are located close to each other, this approach optimizes their scheduling policy. Given the feature of network bandwidth restriction, ESSA adopts an approach to integrate PSO algorithm and SSA, which enable nodes to profit the experience from other nodes. This approach speeds up the learning rate in great deal. The simulation results further demonstrate that ESSA can largely reduce much more energy than S-MAC with duty cycle 10% while the latency for data packs approximates to that of S-MAC with duty cycle 60%. Comparing with SSA, the ESSA can further reduce the energy consumption and latency of data packs at the completion of the simulation test.更多还原