粗糙集与Hamming网络集成的无线传感器网络节点故障诊断

【作者】 代传龙；

【导师】 雷霖；

【作者基本信息】 电子科技大学 ， 检测技术与自动化装置， 2007， 硕士

【摘要】 采用自动配置的无线传感器网络(Wireless Sensor Networks, WSN),是近几年全球范围的新技术研究热点之一。它集成了无线通信技术、传感器技术、微机电系统技术和分布式信息处理技术。它将逻辑上的信息世界与客观上的物理世界融合在一起,改变了人类与自然界的交互方式,在民用和军用等各个领域具有极高的应用价值。WSN已经被认为是极具潜力和影响力的技术之一。尽量延长WSN的使用寿命和保证传递的信息的可靠性分别是WSN的关键技术之一。及时、正确的节点故障诊断可以保证无线传感器网络传递的信息的可靠性,为在上位计算机或中心节点中开展有效的路由规划和节点管理,开展无线传感网络节点的远程维护,延长无线传感器网络的使用寿命起到关键作用。算法融合是近年来国际上流行的一种研究方式。本文提出粗糙集与Hamming神经网络的集成算法来解决能量有限而又具有显著不确定性的WSN节点在线故障诊断问题。首先利用粗糙集理论进行故障诊断决策表的约简,然后用约简后的数据训练神经网络,最后用训练后的Hamming神经网络对WSN节点故障进行诊断和定位。本文算法有效地发扬了粗糙集理论在剔除冗余属性方面的作用,又充分结合了神经网络在并行计算和克服噪声干扰方面的功能。并且,在粗糙集理论约简算法中,本文提出了改进的基于属性重要性的归纳属性约简算法,提高了计算效率;在Hamming神经网络中,本文改进了竞争层网络,避免了竞争层过多的迭代计算。大量仿真实验表明:基于粗糙集与Hamming神经网络集成的故障诊断算法揭示了WSN节点故障特征信息的内在冗余性;能准确快速地解决WSN节点的在线故障诊断问题。在获得的信息不完整或部分信息有误的情况下,也能给出WSN节点的合理故障诊断。与传统的if-then诊断规则相比较,在故障特征数据可靠性降低时,愈加突显出本文算法在诊断正确率上的优势。本文算法有诊断正确率高、通信代价小和能量消耗低的特性。总之,本文算法提高了故障诊断的鲁棒性,增强了能量有限的WSN的实用性。更多还原

【Abstract】 Recently, the research on wireless sensor networks (WSN) with automatic configurations has become a very dynamic area. It is an interdisciplinary study, integrating radio communication, sensor, micro-electro-mechanism system and distributed information process. Combining the logic information world with the external physical world, it can be applied in various fields, and thus nowadays regarded as one of the most promising technologies.Key issues for wireless sensor networks include prolonging the lifetime and assuring the reliability of information transmission. Accurate and timely diagnosis of node fault is essential in ensuring the reliable transmission of information through WSN, facilitating the efficient routing planning, node management for upper computers or sink nodes, long-distance node service and therefore prolonging the lifetime of WSN.Algorithm fusion is a popular international research mode in recent years. In the dissertation, we proposed an algorithm to achieve the online fault diagnosis of WSN nodes with limited energy under major uncertainty. It is an application of the rough set theory and Hamming neural networks. By applying rough set theory, the fault decision-making table is first reduced and then used to training the Hamming neural networks, which is used to diagnose and locate the faults of WSN nodes. The algorithm utilizes the capability of data reduction by rough set theory, and the advantages of Hamming networks in parallel computation and interference resilience. We proposed an improved attribute reduction algorithm based on attribute importance, which improves the computation efficiency. We also improved the competition layer of the Hamming network to avoid too much iterations in computation.Simulation results show that the algorithm reveals the inherent redundancy of fault characteristics in WSN nodes. The algorithm can resolve online fault diagnosis of WSN nodes accurately and timely. It can also yield reasonable diagnosis result when information is incomplete or partial information is false. The proposed algorithm outperforms the traditional if-then diagnosis rules significantly in terms of the accurate diagnosis rate, especially when the fault characteristic data are not reliable. The advantanges of the proposed method include high diagnosis accuracy, low communication cost and low energy consumption. In conclusion, the algorithm has improved the robustness and practicality of fault diagnosis under the limited energy constraint.更多还原