基于RFID的超市购物数据分析算法研究

发布时间：2018-02-07 13:05

本文关键词： 射频识别相位改进的K邻近算法层次聚类算法　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：近几年,非接触的射频识别技术(Radio Frequency Identification,RFID)已成为人们生活中不可或缺的一部分,凭借着自身体积小、远距离通信、无线识别、具有一定存储能力且无需经常人工维护等诸多特点,RFID成为信息数据收集领域的重要部分。随着我国物联网(Internet of Things,IoT)产业的飞速发展,RFID已广泛地应用于包括,管理供应链、跟踪牲畜、防止假冒、门禁系统、自动结帐以及图书馆书籍跟踪等诸多领域。无源RFID系统具有无需内置电池、无线识别、成本低等优势,使其成为商场购物数据分析的重要技术。本文将RFID系统应用于商场购物数据的深度分析,通过对商品各个状态的实时信息采集和分析,挖掘顾客的感兴趣商品和相关的商品,以及商场的热点区域。为商场针对性地进货、促销、以及商店布局提供了科学的理论依据,进而能够根据客户的喜好推荐相关产品,为顾客提供更高质量的服务。但如何在海量标签同时存在,同时移动的情况下,准确、高效完成购物数据的收集和分析是难点问题。现有的多数数据分析算法时延大、能耗大且算法复杂不易实现,都很难高效、可靠且具有针对性地解决商场购物数据准确、深入分析的问题。本文提出的购物数据分析算法,针对性地解决了超市购物数据深入分析问题。首先使用阅读器收集无源RFID标签的相位信息,将收集的相位信息转换为商品的相对移动速度。其次,考虑到密集放置的RFID标签间的相互干扰,针对性地找出了在大型场所中密集放置RFID时的变化规律,并在此基础上对k最邻近算法(k-Nearest Neighbor,kNN)做出改进,提出了改进的k NN算法(Improved k-Nearest Neighbor,I-kNN),利用I-kNN对收集到的相对移动速度序列进行分析,对不同状态商品进行分类。之后,利用层次聚类(Hierarchical Agglomerative Clustering,HAC)算法将训练样本集中的每个数据点都当做一个聚类,通过计算两个聚类之间的距离,不断地将速度相近的商品进行合并,识别出各类别商品的相关性。最后,利用现有的商用设备,对所提出的系统建立原型,并进行了算法的实现和性能评估。结果表明,我们的方法在购物数据分析算法在实际中是可行的,在计算量和时间延迟方面明显优于其他算法。
[Abstract]:In recent years, the contactless RFID technology, Radio Frequency Identification (RFID), has become an indispensable part of people's lives, relying on their small size, long-distance communication, wireless identification. With the rapid development of Internet of things of IoT industry in China, RFID has been widely used in including, managing supply chain, and so on. Tracking livestock, preventing counterfeiting, access control systems, automatic checkout and library book tracking. Passive RFID systems have the advantages of no built-in batteries, wireless identification, low cost, etc. In this paper, the RFID system is applied to the in-depth analysis of shopping data in shopping malls. Through the real-time information collection and analysis of the various states of goods, the paper excavates the goods of interest to customers and related commodities. And the hot spot area of the mall. It provides the scientific theoretical basis for the shopping mall to purchase, promote, and store layout, and then can recommend the relevant products according to the customer's preference. But how to complete the collection and analysis of shopping data accurately and efficiently is a difficult problem in the case of simultaneous existence of mass tags and simultaneous movement. Most existing data analysis algorithms have a long time delay. It is difficult to solve the problem of accurate and in-depth analysis of shopping data reliably and pertinently. This paper solves the problem of in-depth analysis of supermarket shopping data. First, we use readers to collect the phase information of passive RFID tags, and convert the collected phase information into the relative moving speed of goods. Taking into account the interference between densely placed RFID tags, this paper finds out the variation law of RFID in large places, and improves the k-nearest neighbor algorithm (k nearest neighbor NN). In this paper, an improved kNN algorithm is proposed to improve k-nearest neighbor I-kNNNs. By using I-kNN, the collected relative moving velocity series are analyzed, and the goods in different states are classified. The hierarchical Agglomerative clustering algorithm is used to treat every data point in the training sample set as a cluster. By calculating the distance between the two clusters, the items with similar speed are continuously merged. Finally, using the existing commercial equipment, the prototype of the proposed system is established, and the algorithm implementation and performance evaluation are carried out. The results show that, Our method is feasible in the analysis of shopping data and is superior to other algorithms in computation and time delay.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.44

【参考文献】