基于数据挖掘的指纹室内定位

发布时间：2018-03-09 08:17

本文选题：数据挖掘　切入点：指纹定位　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：近些年来,随着移动互联网的迅猛发展以及智能移动终端的广泛普及,基于位置的服务受到越来越多的关注。但受限于室内障碍物较多,空间较为狭小等特点,传统的室外定位算法无法应用于室内定位。目前,基于WIFI系统的指纹法室内定位技术以其部署成本低、组网灵活、易于实现、便于扩展等特点逐渐成为研究的热点。基于WIFI的指纹法室内定位通常选择在定位点上接收到的各个接入点(Access Point,AP)的接收信号强度指示(Received Signal Strength Indicator,RSSI)作为定位的特征指纹,利用RSSI与地理位置之间特殊的映射关系实现定位。然而RSSI容易受到多径、衰减和环境变化的影响,从而导致信号强度测量数据难以构建一个可信的模型,这是提高室内定位精度所面临的主要挑战,需要利用不断发展的新技术加以改进。为此,本文通过对相关文献的查阅,将数据挖掘理论引入到室内定位中,结合实地的数据采集和分析,进行了如下的研究工作:(1)为分析RSSI作为定位特征指纹所表现出来的特性,在WIFI网络环境中实地采集RSSI样本,采用理论与实验相结合的方法,对环境中各个AP的RSSI进行验证。发现WIFI网络中不同AP所发送信号的RSSI具有不确定性和重复性,为了准确描绘RSSI与地理位置之间的关系,需要尽可能多的收集不同AP的RSSI,这会导致定位算法计算量的增加。此外,在室内环境中,为了保证WIFI网络的覆盖和数据传输质量,往往会重复部署很多AP,这些AP之间的RSSI具有很高的重复性。为此,本文提出一种基于主成分分析的指纹降维算法,该算法利用特征空间基变换的原理,将原始高维指纹数据映射到低维,在降低数据维度的同时,去除了不同AP间的冗余信息。经验证,该算法降低了算法复杂度,提高了定位效率。(2)为了提高定位算法的精度,本文提出了一种基于k层网格参数寻优的支持向量回归定位模型。针对RSSI的非线性特征,目前的解决方法是使用基于核函数的支持向量机来构建定位模型。但是支持向量机的定位效果受参数的影响较大,传统的参数寻优算法效率低,耗时大。为此,通过分析传统算法效率不佳的原因,采用分层的思想,在不同参数区间选择不同的搜索步长。经验证,本文提出的k层网格参数寻优算法在计算效率上取得了显著地提升。(3)最后,本文将指纹降维算法和最优参数的支持向量回归定位模型相结合。使用实地采集的数据进行算法仿真。实验结果证明,相较于其他基于数据挖掘的定位算法如传统支持向量机定位算法、KNN定位算法、神经网络算法,本文提出的算法在定位精度方面,表现出优越的性能。
[Abstract]:In recent years, with the rapid development of mobile Internet and the widespread popularity of intelligent mobile terminals, location-based services have attracted more and more attention. The traditional outdoor location algorithm can not be applied to indoor localization. At present, the fingerprint indoor location technology based on WIFI system has the advantages of low deployment cost, flexible networking and easy implementation. The characteristics such as easy to expand and so on have gradually become the research hotspot. The fingerprint method based on WIFI usually selects the received signal strength indication of received Signal Strength indicator (RSSI) from each access point received at the location point as the fingerprint feature of the location. Using the special mapping relationship between RSSI and geographical location, RSSI is easy to be affected by multipath, attenuation and environmental changes, which makes it difficult to build a credible model of signal strength measurement data. This is the main challenge to improve the accuracy of indoor positioning, which needs to be improved by using the new technology. Therefore, this paper introduces the theory of data mining into indoor positioning by consulting related documents. Combined with data acquisition and analysis in the field, the following research work was carried out: 1) in order to analyze the characteristics of RSSI as a location feature fingerprint, RSSI samples were collected in the WIFI network environment, and the method of combining theory with experiment was adopted. The RSSI of each AP in the environment is verified. It is found that the RSSI of different AP signals in WIFI network is uncertain and repetitive. In order to accurately describe the relationship between RSSI and geographical location, The need to collect as many different AP RSSIs as possible, which can lead to an increase in the computation of the location algorithm. In addition, in an indoor environment, in order to ensure the coverage of WIFI networks and the quality of data transmission, Many APs are often repeatedly deployed, and the RSSI between these APs is highly repeatable. In this paper, a fingerprint dimensionality reduction algorithm based on principal component analysis (PCA) is proposed, which utilizes the principle of feature space basis transform. The original high-dimensional fingerprint data is mapped to low-dimensional fingerprint data, and the redundant information between different AP is removed while reducing the data dimension. It is proved that the algorithm reduces the complexity of the algorithm and improves the localization efficiency. In this paper, a support vector regression location model based on k-layer grid parameter optimization is proposed. The current solution is to use kernel function based support vector machine (SVM) to construct localization model. However, the localization effect of SVM is greatly affected by parameters, and the traditional parameter optimization algorithm is inefficient and time-consuming. By analyzing the reasons for the inefficiency of the traditional algorithm and adopting the idea of stratification, different search steps are selected in different parameter intervals. The k-layer mesh parameter optimization algorithm presented in this paper has achieved a significant increase in computational efficiency. In this paper, the dimensionality reduction algorithm of fingerprint is combined with the support vector regression localization model of optimal parameters. The algorithm is simulated using the data collected in the field. The experimental results show that, Compared with other localization algorithms based on data mining, such as traditional support vector machine (SVM) localization algorithm and neural network algorithm, the algorithm presented in this paper shows superior performance in terms of location accuracy.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN92

【引证文献】