【作者】 陈分雄；

【导师】 王典洪；

【作者基本信息】 中国地质大学 ， 地球探测与信息技术， 2013， 博士

【摘要】 无线传感网(Wireless Sensor Networks,简称WSN)是未来网络发展的主流形式,并已成为本世纪一个新科学研究领域。在基础理论和工程技术两个层面提出了许多急需解决的问题。无线传感网成本低廉、低功耗、大规模自组网；传感器节点体积小巧、电池供电、部署灵活；以及能够适应监测人力难以到达的恶劣环境；这些特点使得无线传感网极大地提升了灾害预防的监控能力。为了及时监测各种可能发生的突发事件(如山体滑坡、大气污染、森林火灾等),必须关注传感器节点采集到的异常测量值。因此,实时准确地检出异常数据,并预警特定事件,具有十分重要的意义。无线传感网的异常事件检测技术概括起来主要分为两类：1)点异常检测方法。点异常即如果传感数据超过设置的某个阂值,则认为事件发生。这种方法只适合小规模、短期的单一事件监测任务。2)模式异常检测方法。在一些长期渐变环境监测中,突发性的复杂事件往往很难由指定属性阈值的超限进行报警,不能用简单的闽值来描述,但可以看做一种模式(事件模式),因可采用模式识别技术进行异常检测。目前,大部分模式异常检测方法都是在原始采集数据空间上进行,即不对传感器节点采集的数据进行任何变换,虽然这种方法的有一定检测精度。但算法计算量大、容错性差,节能效果有限。能否在经过压缩处理后的数据空间上进行异常检测?并且,无线传感网中异常事件检测技术还要面临两个主要挑战：1)检测精度。由于受环境噪声和网络中各种故障的影响,传感器节点经常给出错误的监测值,这势必会影响到异常事件检测的可靠性。因此,检测方法必须具有容错性。2)能量有效性。传感器节点具有非常有限的能量储备,无线传感网事件监测的网络生命期取决于节点能耗,因而检测方法必须具有节能性。本文以无线传感网中事件监测信息的压缩提取与异常检测为基本对象,研究无线传感网的数据压缩与融合,以及事件监测的表示模型与异常检测方法。以期突破大规模长期部署的无线传感网事件监测技术瓶颈,探索高效节能的轻量级数据压缩算法；寻求在压缩数据空间上快速准确地进行异常事件检测算法,并能够量化异常的程度；设计高精度、快速地信号重构算法,以恢复原始数据中的异常信息。本文在数据压缩与融合算法方面分别提出了基于线性分段的数据压缩方法ETEO、基于WSN分簇路由协议的多重主成分分析数据压缩算法、基于压缩感知理论和ETEO线性分段的二次混合数据压缩方法,以及基于改进的符号聚合近似数据压缩方法。这四种压缩方法本质上都利用了无线传感网监测数据的空时相关性。在事件监测的表示模型与异常检测方面,针对事件异常阈值已知的单一事件监测,提出一种WSN监控异常事件的容错快速检测算法。针对异常发生机理未知的复杂事件监测,在前述的数据压缩方法基础上,分别从数据分段线性压缩、压缩感知与分段线性混合压缩、改进的符号聚合近似压缩后的模式空间设计了三种对应的异常检测算法。同时,提出一种基于凸优化的L1范数快速求解方法。在改进的聚合近似FSAX符号模式空间上,借用生物信息学的混沌图来可视化地刻画事件的空时动态变化信息。论文最后给出了基于无线传感网的水质监测平台实例,采用最新的WSN协议和硬件平台设计了多参数水质信息采集系统,并进行了实验验证。本文的贡献主要包括以下几个方面：1)提出基于WSN节点级ETEO数据压缩与异常事件检测算法。该方法结合图像处理中边缘算子的基本思想,通过提取能表征节点数据序列趋势的边缘点,将其简化为近似线性分段,不需要预定义与数据压缩有关的阈值,通过实验比较了ETEO算法的压缩性能。在此基础之上,又给出一种基于ETEO模式的WSN异常检测方法,以模式的本地异常因子来表征数据序列的异常程度,而不是在原始数据序列上对单个数据点进行异常检测,提高了WSN事件检测效率和准确性。2)提出基于WSN网络级的多重主成分分析数据压缩方法。结合WSN分簇路由协议在多层路由上迭代使用主成分分析,对节点采集的数据矩阵进行压缩处理,以消除在一段时间同一簇内不同节点所采集数据间的时空相关性,同时,消除了同层路由上相邻簇首提取主成分间的空间相关性,提高了WSN数据压缩性能,有效地消除了数据空时冗余。3)针对异常阈值已知的事件监测,提出一种WSN监控异常事件的容错快速检测算法。该算法采用滑动滤波窗口纠正由瞬时性故障引起的错误节点监测值,通过节点的置信水平自适应地消除由永久性故障引起的错误节点监测值。同时,该算法采用一种滑动窗口匹配机制来检测节点监测数据的变化趋势,对节点能否检测到事件进行预测,提高了WSN检测异常事件的及时性和可靠性。4)为了解决评价WSN协议和算法性能时缺乏大规模的测试数据源问题,提出一种基于半变差函数和概率统计的空间相关性模型。通过灵活地控制模型的参数,可以合成具有任意空间相关性强弱的大规模数据集：也可以从少量的真实监测数据中快速地抽取参数,生成与之空间相关性匹配的数据集。该模型具有较少的约束条件和广泛的实用性。5)提出基于压缩感知理论和ETEO方法的二次混合压缩与异常检测算法。针对WSN监测对象具有宽动态范围、低信噪比的异常前兆信号识别问题。通过采用光滑化逼近和加速技巧,将经典的凸优化方法引入到压缩感知理论中,提出一种基于凸优化的L1范数快速信号恢复算法。可以准确定位和高精度地重构原信号中包含的异常信息。6)针对WSN监测复杂事件异常变化难以进行精确的数学建模,提出一种基于FSAX-MARKOV模型的异常事件检测算法。并采用混沌表示的可视化模型刻画事件的空时动态变化信息,该模型单纯由监测数据驱动,避免了复杂的数学建模,是一种轻量级的数据压缩与异常检测算法。更多还原

【Abstract】 Wireless Sensor Networks (WSNs) are often deployed for the purpose of detecting significant events or anomalies in the monitored phenomenon, process or structure (henceforth, landslides, air pollution, etc.). Such reactive systems typically collect and process sensor observations to programmatically classify the real world state of the monitored object into one or more classes and take the necessary actions accordingly. For example, in a structural health monitoring system, a building is monitored for structural faults by comparing’a seismic response signal to known stress patterns.Matching or detecting patterns in sensor observations is a common requirement in a number of domains yet the problem of computationally efficient approaches has attracted less attention in comparison with research in network layer protocols. Moreover, solutions are often based on transmitting observations outside the network or to a tier of high capability devices for processing.In this thesis, we assume a homogeneous WSN comprised of resource limited devices and we attempt to solve the problem of pattern matching and detection inside the network. Apart from the ubiquity of the problem, we are motivated by the benefit of an in-network solution, namely prolonged lifetime resulting from reduction of radio communication.We target the extremely resource constrained end of the WSN spectrum that comprises nodes. The constraining factors that differentiate such nodes from other distributed systems are:1) Limited power resources. Typically, nodes are powered by batteries which limit their useful lifetime and specify an energy budget that, in most applications, must be extended as much as possible. Radio communication, sensing and processing share this budget and pose a challenge to developers who must serve the application’s purpose and, at the same time, maximize node lifetime.2) Restricted functionality. Embedded microcontrollers limited RAM and in the vast majority of cases Costly radio communication and limited bandwidth. The fabric that inter connects nodes in a WSN is also the most expensive component with respect to power draw. Minimizing the amount and range of communications, can prolong the lifetime of a WSN. As a rule of thumb, a bit of data transmitted by radio can cost as much as executing1000CPU instructions.Further to the above constraints, WSN application designers are challenged by limited support for software development and a tight coupling between application and system layers.This paper has carried comprehensive summary of event monitoring method includes point exception and mode anomaly detection.1) Point exception if the sensor data exceeds a set threshold; however this method only applies to a single event and combinations of events monitoring tasks. In some long gradient environmental monitoring, sudden complex events is often difficult to be overrun by the specified attribute threshold alarm can not be described using a simple threshold, but can be seen as a pattern may be mode recognition technology to anomaly detection.2) Mode anomaly detection method can be divided into the original data space and the compressed data space anomaly identification. Currently, most of the abnormal pattern detection methods are in the raw data space, i.e. not the sensor nodes collect data for any transformation, although this method has high detection accuracy. However, the computational complexity of the algorithm, fault tolerance and energy-saving effect is limited. Is it possible exception information extraction and recognition after data compression processing space?This paper attempts to three steps to solve this problem:1) explore energy efficient lightweight data compression algorithms;2) seek abnormal event detection algorithms to quickly and accurately in the compressed data space, and be able to quantify the degree of abnormality;3) design of high-precision, quickly signal reconstruction algorithm, to restore the original data of the abnormality information. To this end, the paper from two aspects of in-depth research: space-time data sequence compressed representation model and integration, event monitoring and anomaly identification, event monitoring technology for large-scale long-term deployment of wireless sensor networks to provide a series of new approaches.The work described in this thesis makes contributions that address the WSN application issue of pattern matching and detection, and offers a computationally efficient implementation of reactive functionality. Moreover, it limits radio communication and MCU active time in order to complement the generic goal of prolonged WSN lifetime. Specifically, we make the following contributions:1) Provides the abnormal event detection algorithm based on the ETEO data compression pattern. By the ETEO method compression the scries into patterns set and extraction its features, series will be mapped to the pattern feature space, then using pattern recognition judgment whether the abnormal events occur.2) Propose an algorithm of data compression based on multiple Principal Component Analysis (multiplc-PCA), iteratively using PCA method in multiple layers.3) For the monitoring of the anomaly threshold known single event, in the real-world applications, sensor nodes may be invalid and work abnormally because of the impact of environmental factors or faults. In order to improve the reliability of event detection, a distributed fault-tolerant abnormal event detection scheme, which utilizes temporal and spatial correlation to detect and correct faults, is proposed in this thesis. Confidence level of sensor nodes is used to manage and adjust sensor nodes’ status, resulting in the isolation of the invalid nodes from the network and the decreasing of invalid nodes’ influence on event detection. In addition, the scheme utilizes a sliding window match to detect the trends of sensor data and predict whether the nodes can detect the events to reduce the response time of event detection.4) Spatial correlation of sensor networks has great use in data aggregation, transmission, encoding, data compression and other important applications. This thesis aims to discuss the spatial correlation of the sensor network, which mainly includes giving a general model to describe the spatial correlation, using the model inferring the relationship of parameters, and giving the way of calculation or estimation the parameters, then we give a hypothesis testing the parameters result obtained from the original data set.5) Put forward the secondary hybrid based CS theory and ETEO methods compression anomaly detection algorithm. The algorithm greatly reduces the amount of calculation of the abnormality detecting process; for abnormal precursor signals with a wide dynamic range, low signal-to-noise ratio, the proposed new fast solution based on the L1norm convex optimization method can quickly and accurately reconstruct the original signal.6) Present an improved symbolic aggregate approximation method to build the energy-efficient space-time data compression and fusion mechanism and the chaos representation model of WSN event monitoring to depict the space-time change of the complex event information, by using the statistics and fractal characteristics of the FSAX-MARKOV model.We believe that the above contributions provide a competitive pattern matching and detection family of algorithms that can be used in a variety of reactive WSN applications.更多还原