基于联合神经网络的WSN节点和网络故障诊断研究

发布时间：2018-02-23 16:48

  本文关键词： WSN节点和网络故障 联合神经网络 故障征兆信号 径向基Elman神经网络 双参数实数编码　出处：《电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：近年来,物联网被视为继计算机、互联网和移动通信之后的又一项信息产业的革命性技术而得到广泛重视。作为物联网的一种实现形式之一,无线传感器网络(Wireless Sensor Networks,WSN)显示出良好的应用前景。实际应用过程中,WSN往往工作在复杂、恶劣的环境中,很容易因受到干扰、损坏而出现故障,从而严重影响其工作效率和质量。同时,因其工作环境的限制及自身的技术特点,这些故障很难人为亲自加以排除,诊断技术也有别与传统网络。因此,节点及网络故障诊断与容错是无线传感器网络技术研究的重要内容之一。根据无线传感器网络的结构、功能特性,结合以有故障诊断方法,针对无线传感器网络中可能出现的节点故障和网络故障,本文合理提取出用于节点故障诊断和网络故障诊断的多个故障征兆信号,并在此基础上提出了一种由两级功能不同神经网络组成联合神经网络的无线传感器网络节点和网络故障诊断方案。第一级神经网络作为预测器,用于预测节点传感器的输出,检测WSN节点传感器的故障,以此产生传感器单元故障征兆信号。在该预测器实现上,对Elman神经网络进行了改进,提出了一种径向基Elman神经网络结构,并对其训练算法进行了推导。然后建立了传感器故障仿真模型,对基于径向基Elman神经网络传感器单元故障检测进行了仿真验证。第二级神经网络作为分类器,用于对所有故障征兆信号进行模式分类,以实现对WSN可能的故障的类型进行判别,该神经网络采用RBF神经网络实现。为了提升该网络的训练效率,对量子遗传算法进行了深入研究,提出了一种量子遗传算法的双参数实数编码方式,并结合现有的混合递阶遗传算法,提出了一种混合递阶量子遗传算法,用于RBF分类器的学习。该故障诊断方案的验证通过计算机仿真与实物实验相结合的方式进行,先在计算机上对故障诊断方法中涉及到的各项技术进行逐一仿真验证,而后整体在无线传感器网络实物上进行测试验证。仿真、实验表明,该无线传感器网络故障诊断方法,能同时对无线传感器网络的节点级与网络级故障进行诊断,并具有较高的故障检测率,满足实际应用的基本要求。本文在理论上的探索,将对神经网络和进化算法的进一步发展起到一定的参考作用。
[Abstract]:In recent years, the Internet of things has received extensive attention as a revolutionary technology in the information industry after computers, the Internet of things and mobile communications. Wireless Sensor Networks (WSNs) show good application prospects. In practical applications, WSNs often work in complex and harsh environments, and are prone to malfunction due to interference and damage. Therefore, due to the limitations of its working environment and its own technical characteristics, these faults are difficult to be personally eliminated and the diagnostic techniques are different from those of traditional networks. Node and network fault diagnosis and fault tolerance is one of the important contents of wireless sensor network technology. According to the structure and function of wireless sensor network, combining with the method of fault diagnosis, Aiming at the possible node faults and network failures in wireless sensor networks, this paper reasonably extracts multiple fault symptom signals for node fault diagnosis and network fault diagnosis. On the basis of this, a scheme of node and network fault diagnosis of wireless sensor network composed of two-level neural networks with different functions is proposed. The first stage neural network is used as predictor to predict the output of node sensor. The fault of WSN node sensor is detected to produce the signal of fault symptom of sensor unit. In the realization of the predictor, the Elman neural network is improved, and a radial basis function (Elman) neural network structure is proposed. Then the sensor fault simulation model is established, and the fault detection of sensor unit based on radial basis function (Elman) neural network is simulated and verified. The second stage neural network is used as classifier. In order to improve the training efficiency of WSN, the neural network is implemented by RBF neural network, which is used to classify the patterns of all fault symptom signals in order to distinguish the possible fault types of WSN. In this paper, quantum genetic algorithm (QGA) is studied in detail. A quantum genetic algorithm (QGA) with two parameters is proposed, and a hybrid hierarchical quantum genetic algorithm (HQGA) is proposed in combination with the existing hybrid hierarchical genetic algorithm (HGA). It is used in the learning of RBF classifier. The method of fault diagnosis is verified by computer simulation and physical experiment. The simulation results show that the method can diagnose both node level and network level fault of wireless sensor network. It has high fault detection rate and meets the basic requirements of practical application. The theoretical exploration in this paper will play a certain reference role in the further development of neural network and evolutionary algorithm.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP212.9;TN929.5;TP183
，

本文编号：1526881