基于最小二乘和支持向量机的WSN节点定位方法研究

【作者】 肖勇；

【导师】 何学文；

【作者基本信息】 江西理工大学 ， 机械电子工程， 2011， 硕士

【摘要】 无线传感器网络(WSN)作为一种新兴的分布式测控网络技术,被认为是二十一世纪改变世界的十大新技术之一,具有非常广阔的应用前景。定位是无线传感器网络的重要支撑技术之一,传感器节点实现自定位是提供监测目标位置信息的前提。实现高效、可靠、准确的节点定位对目标跟踪及提高路由效率具有重要的意义,是目前无线传感器网络应用研究中迫切需要解决的关键技术问题。目前普遍研究的节点定位算法一般是在二维环境下实现的,与实际三维环境下的应用还有较大的差距。因此,三维节点定位技术成为了目前无线传感器网络研究的热点和前沿课题。最小二乘定位算法此前一直是作为经典的二维节点定位算法,而把最小二乘二维定位算法扩展到三维定位具有重大意义和应用价值。本文结合国家自然科学基金项目的研究内容,重点研究了无线传感器网络的节点定位技术。在研究了经典的最小二乘二维节点定位算法的基础上,把它扩展到了三维,并且针对经典算法易受测距误差影响的不足,结合支持向量回归机较强的学习能力和泛化能力的优点,研究了基于最小二乘支持向量机的三维节点定位算法,大大提高了在测距误差较大的三维环境下节点的定位精度。本文所作的主要研究工作如下:1、查阅了大量相关文献后,综述了国内外无线传感器网络,尤其是节点定位技术的研究现状,总结出了几种主要的无线传感器网络定位技术的特点和分类,并根据这些不同定位技术的分类及特点,选用了基于测距的含锚节点定位方法作为本文的研究方向。2、分析了无线传感器网络节点定位技术的原理,探讨了定位算法全方位的六个评价指标,它们分别是:定位误差、锚节点个数、适用规模、节点密度、功耗、容错性。其中定位误差是大多数定位技术的首要评价指标,降低定位误差提高定位精度是研究各种新型节点定位算法的重要目标,本文正是为了改善最小二乘三维节点定位算法的定位误差提出了一种新的基于最小二乘支持向量机的三维节点定位算法。3、以经典的最小二乘二维定位算法为研究对象,研究了最小二乘二维节点定位的原理,数学模型,并从最小二乘二维节点定位的数学模型着手将其从二维扩展到了三维。在Matlab 7.0仿真软件下编写了最小二乘二维和三维节点定位算法,并分别进行了节点定位的仿真实验,验证了最小二乘定位理论从二维向三维扩展的可行性。4、针对最小二乘定位算法容易受节点测距误差影响、在不同的环境下适应性差的不足,提出了一种基于最小二乘支持向量机的三维定位算法。特别地在对本算法的核函数参数、定位区域的网格划分宽度的寻优过程进行大量的研究和实验,提出了一种结合“实验经验的穷举法”求参方法,借助软件Matlab 7.0进行了算法的仿真分析。仿真结果表明,所求参数使最小二乘支持向量机的三维节点定位算法达到了良好的定位精度。5、在理论分析和仿真分析基础上,进行了真实环境下的无线传感器网络定位系统的实验研究。对最小二乘支持向量机三维节点定位算法进行实验,得到了低于10%的定位误差率,表明本算法有良好的实际效果和较高的实用价值。6、为了论证最小二乘支持向量机三维节点定位算法相对于最小二乘三维节点定位算法的优越性,对两种算法进行了定位误差回归分析和比较分析。回归分析表明:后者的定位误差基本跟节点测距误差成线性关系,而前者的定位误差跟节点测距误差成非线性关系。比较分析表明:在较低的测距误差环境下两种定位算法的定位误差相近;但是在测距误差超过16%的环境中,新的算法能大大提高定位精度,相比传统的最小二乘定位算法平均能降低25%的定位误差,甚至最高能降低50%的定位误差。综上所述,论文把理论研究、仿真分析和实验验证相结合对无线传感器网络的节点自定位技术进行了深入的研究,提出的新的基于最小二乘支持向量机的三维节点定位算法具有较强的环境适应性,在测距误差较大的三维环境下能够大大降低节点定位误差,为无线传感器网络节点定位技术提供了一种新的方法,对无线传感器网络技术的发展和应用进行了有益的探索和尝试。更多还原

【Abstract】 Wireless Sensor Network(WSN), a new distributional observation network, was considered that is one of the ten new technologies in the 21st century,and it has the very broad application prospect. The nodes localization technology is one of the most supporting technologies in WSN. In order to provide the location information, sensor nodes should achieve self-localization first. It’s very significant to achieve highly effective, reliable and accurate node localization for target tracking and route efficiency, and thus that also is the key technology to be solved in the present wireless sensor network. Currently, the popular study of nodes localization algorithm is generally implemented in the two-dimensional environment, which is lag far behind the actual application in three-dimensional environment. Therefore, three-dimensional nodes localization technology becomes the current hot spot and cutting-edge issue for WSN research. The least-square localization algorithm has been a classic two-dimensional nodes localization algorithm before. An extending to three-dimensional localization algorithm will be of great significance and application value. Combined with National Natural Science Foundation research project, this paper aims at the nodes localization technology in WSN. There will be a brief study of the typical least-square two-dimensional nodes localization algorithm, based on which an extending to three-dimensional localization algorithm will be worked out. And then, because the classical least-square algorithm is easily influenced by ranging errors between Nodes, which is a shortcoming, combined with strong learning ability and generalization ability of Support Vector Machines for Regression (SVR), this paper makes a research on LSSVR three-dimensional nodes localization algorithm (LSSVR3DNL). It greatly improved localization accuracy of nodes under three-dimensional environment with large ranging error, which will be significant to the application of WSN in the three-dimensional environment.Specifically, the paper general goes in this way:1. After consulting a great deal of relevant literature and analyzing the research situation and prospects of WSN, several major classifications of WSN localization technology and characteristics are summarized. In addition, depending on these different classifications and characteristics of the localization technology, this paper selects the anchor nodes localization method based on ranging as the direction of this research, laying foundations for later study.2. After analyzing the principles of nodes localization technology in WSN, it has made some discussion about six evaluation indicators of the localization algorithm. Including: localization error, the number of anchor nodes, nodes density, power consumption, fault tolerance. Among that localization error is the primary evaluation indicator of most positioning technologies. Reducing the positioning error to improve the positioning accuracy is an important goal to study a variety of new localization algorithms. This paper proposes a new three-dimensional nodes localization algorithm that based on LSSVR aiming at reduce ranging error caused by the least-square three-dimensional nodes localization algorithm.3. Taking the classical least-square two-dimensional localization algorithm as its research object, this article studies the principles of the least-square two-dimensional nodes localization and mathematical model. What’s more, the two-dimensional is extended to three-dimensional proceeding from the mathematical model of least-square two-dimensional nodes localization. With the help of Matlab 7.0 simulation software, the least-square two-dimensional and three-dimensional nodes localization algorithms were programmed. After nodes localization simulations, it verifies the feasibility of extending two-dimensional to three-dimensional for the least-square localization theory.4. A three-dimensional localization algorithm based on LSSVR is proposed for the situation that the classical least-square algorithm is easily influenced by ranging error between nodes and its poor adaptability in different environments. A great deal of researches has been carried out for the optimizing process of parameters of kernel function, mesh width of oriented area, which are contained in LSSVR3DNL. It proposes method of exhaustion which is combined with practical experience, using the Matlab 7.0 simulation the algorithm. Simulation results show that the parameters obtained by this method make the LSSVR3DNL achieve high localization accuracy.5. The wireless sensor network positioning experiment in real environment was made after doing lots of theoretical analysis and simulation analysis, and conducted an experiment for LSSVR3DNL. A localization error rate less than 10% was gotten. This algorithm shows a good practical effects and value.6. This paper provides regression analysis and comparison on these two kinds of nodes localization algorithms in order to demonstrate the superiority of LSSVR3DNL over LS3DNL. The regression analysis indicates that the localization error of LS3DNL is more or less linear with nodes ranging error, while the relationship between the localization error of LSSVR3DNL and nodes ranging error is nonlinear. The comparison shows that the localization errors obtained by the two kinds of localization algorithms under the environment of small nodes ranging error are similar. However, when the nodes ranging error is over 16%, LSSVR3DNL can greatly improve localization accuracy with an average reduction of 25% of localization error, sometimes even to an utmost of 50%, compared with the traditional LS localization algorithm.To sum up, the paper integrates the theoretical study, simulation analysis and experimental validation to make a deep research on nodes self-localization technology in WSN. The new LSSVR3DNL proposed in this paper has strong adaptability to the environment. Nodes localization errors can be greatly reduced under three-dimensional environment with large ranging errors. It provides a new method for WSN nodes localization technology, which is a kind of useful exploration to the development of WSN.更多还原