【作者】 张明华；

【导师】 张申生；

【作者基本信息】 上海交通大学 ， 计算机应用技术， 2009， 博士

【摘要】 随着无线网络的普及和普适计算技术的发展,基于位置的服务受到越来越多的关注,在紧急救助、医疗保健、个性化信息传递等领域显示出巨大的活力。其中,定位技术是实现普适计算应用和基于位置的服务的必要前提和关键。但是,现有的定位技术,特别是室内定位技术,在使用成本、应用范围和便携性等方面还不能满足普适计算应用的需要,限制了基于位置的服务在用户中的普及。基于WLAN和位置指纹的室内定位技术由于使用范围广,能够以纯软件的方式实现,定位系统成本低等优点,成为近年来普适计算和位置感知领域的一个研究热点。本文对这一技术进行了较深入、系统的研究,并针对定位准确度、移动设备能耗方面存在的问题,采用人工智能和数据挖掘理论,提出了相应的解决方案。通过算法比较和实验分析,证明了方案的有效性和可行性。论文的主要工作和创新点如下:(1)通过在实际WLAN环境中测得的位置指纹数据,从定位的角度分析了无线信号在室内的传播特性,并提出一个表示定位平均误差的数学模型。在基于WLAN和位置指纹的定位技术里,采样点间距、接入点的个数以及环境的干扰是影响定位平均误差的重要因素,但是对于这些参数的选取目前还没有较系统的指导,主要依靠经验来确定。该数学模型形式化地概括了定位问题中的一些关键因素。通过仿真实验,本文分析了各因素与定位平均误差之间的关系,为设计定位算法、部署定位系统提供了一定的理论支持。(2)根据多源信息融合的思想,提出了一种基于Dempster-Shafer证据理论的定位方法。由于受室内复杂环境里噪声的干扰,位置指纹数据中常含有不确定性因素,提高定位准确度是当前研究的重点之一。该方法将接入点作为提供接收信号强度数据的信息源,能够为不同的信息源赋予不同的信任度,详细地描述和区分了不同接入点对定位结果的贡献能力。在选择信息源时,还采用了本文所提出的一种接入点选择和匹配方法——最大匹配法。与目前常用的一些定位方法相比,本文所提的方法能够更加有效地估计用户的位置,准确度更高。(3)提出了一种基于高斯混合模型(GMM)的位置指纹聚类算法。鉴于越来越多的用户使用主要依靠电池供电的移动设备,如何减小定位算法的计算量,节省电能也是一个十分重要的问题。本文提出的算法用高斯混合模型表示位置指纹的簇,通过考虑接收信号强度的概率分布,克服了现有聚类算法只考虑接收信号强度大小的相似性的不足,从而减小聚类对于含噪声数据的敏感性。此外,该算法中的参数含义明确,易于使用。实验结果表明,该聚类算法能更好地降低定位算法的计算量,减小移动设备的能耗。(4)本文以普适计算应用为背景,设计并实现了一个基于WLAN和位置指纹的室内定位系统原型,并在实际的无线局域网环境里对上述方法进行了实验。实验结果表明,本文提出的方案在提高定位准确度的同时,还能够降低定位算法的计算量,增强了定位系统的实用性。更多还原

【Abstract】 With the widespread of wireless networks and development of pervasive computing, location based services (LBS) have attracted more and more attention and shown great energy in many applications, such as emergency, medical care and customized information delivery. Location estimation is a prerequisite and key to implement pervasive computing applications and LBS.However, existing location estimation technology, especially in the indoor environments, cannot satisfy the needs of pervasive computing applications in system cost, application area and portability, which limits the popularization of LBS in common people. In recent years, the indoor location estimation technology based on WLAN and location fingerprinting has become a hot research topic in the field of pervasive computing and location awareness for the advantages that it works in a more wide area, can be implemented simply in software and doesn’t cost too much.Aiming at existing problems and requirements in location estimation accuracy and energy consumption of mobile device, the thesis makes a deep and systematic study, and proposes corresponding solutions using theory of artificial intelligence and data mining. Through algorithm comparison and experimental analysis, the validity and practicability of the proposed solutions are demonstrated. The major contributions of this dissertation are:(1) With the location fingerprints measured in a practical WLAN, the thesis analyzes some characteristics of radio signal propagation in indoor environments from positioning point of view. Based on this, a mathematical model of average error of localization is proposed.In WLAN and location fingerprint based location estimation technique, the sampling grid spacing, the number of access points, and the interference of environment are key factors for average error of location estimation. However, there isn’t a general strategy on how to determine these parameters. They are mostly determined by experience. The mathematical model formally describes the key factors for errors in fingerprint based location estimation. Through emulation experiments, relationships between the average error and the number of access points, the size of sampling grid and environmental factors are analyzed, which helps to design location estimation algorithm and deploy location estimation system.(2) According to the idea of multi-source information fusion, the thesis presents a location estimation method based on Dempster-Shafer evidence theory.Due to the interference from noises of complex indoor environments, data of location fingerprints always contain uncertainty. Improving location estimation accuracy is one of important research focuses. The method considers access points as information sources of received signal strength, and assigns different belief to information sources. The contribution of access points to location estimation results is amply described and differentiated. To choose the information sources, an access point selection method proposed by the thesis, that is, the maximum matching method is used. Compared with existing location estimation methods that often used, the proposed method can estimate user location more efficiently and get higher location estimation accuracy.(3) This thesis proposes a Gauss mixture model (GMM) based location fingerprint clustering algorithm.As more and more clients choose small, self-maintained devices which heavily depend on battery power, how to reduce computation cost in location estimation and save energy is also a very important problem. The proposed clustering algorithm uses a Gauss mixture model to represent location fingerprint clusters. By taking account of the probability distribution characteristic of received signal strength, the algorithm overcomes the shortcoming of other location fingerprint clustering algorithms which only consider the similarity of signal strength values, and thus alleviates the sensitivity to noisy data. Moreover, parameters in the algorithm are easy to determine. Experimental results show that the clustering algorithm can effectively reduce computation cost of location estimation algorithm and decrease energy consumption of mobile devices.(4) Oriented to pervasive computing applications, a prototype of indoor location estimation system based on WLAN and location fingerprinting technique is designed and implemented. All experiments are conducted in a practical indoor WLAN environment. Experimental results show that the proposed solutions not only improve location estimation accuracy but also reduce computation cost so that the practicability of location estimation system is enhanced.更多还原