无约束不确定RFID数据流近似去噪

RFID technology has been the key technology in the field of Things of Internet, and it appears enormous development potential and extensive application perspective. However, the data from the RFID readers are always imperfect and full of noise due to the instability and influence easily by external environment of radio frequency technology, which has been the important factor blocking the application widely of RFID technology currently. In order to enhance the integrity and reliability of RFID data stream, aimed at three typical kinds of noise such as missing-readings, cross-readings and duplicate readings, taking sufficiently the features like uncertainty, randomness and proximity into account, this project mainly researches approximative denoising theories and technologies in RFID data stream without using any constraints. The research contents of this project mainly include: the model of uncertain RFID data streams and the framework of the denoising system; the model of damping sliding time windows and the adaptive window adjustment strategy; the approximate duplication elimination strategy based on a probability comprehensive Bloom filter; the strategy of fake elimination based on the probability kernel density for the cross-readings, and the strategy based on Top-k probability frequency pattern mining for the missing-readings. The main characteristics of this project are: establishing a unified denoising system framework with interrelation and mutual complementation based on the probability, and presenting approximate denoising methods and theories using uncertain data stream mining methods without using constraint conditions and priori knowledge. Therefore, the research results of this project have a wide applicable range.

RFID技术已成为物联网领域不可或缺的支撑技术，正显现出巨大的发展潜力与广泛的应用前景。然而，由于射频技术自身的不稳定及易受外部环境影响，来自读写器的数据通常是不完整且具有噪声，已严重阻碍到RFID技术的推广应用。为提高RFID数据流的完整性和可靠性，本课题根据RFID数据流的不确定性、随机性及近似性等特点，针对"重复读"、"交叉读"和"漏读"等三类典型噪声，研究无约束RFID数据流近似去噪理论与技术，主要研究内容包括：不确定RFID数据流模型及去噪系统框架；衰减滑动时间窗口模型及窗口自适应调整策略；基于概率综合Bloom过滤器的近似消重策略；基于概率核密度的"交叉读"去伪策略；基于Top-K概率频繁模式挖掘的漏读填补策略等。本项目主要特色为：建立相互联系、相互补充、基于概率的统一去噪系统框架，以及基于不确定数据流挖掘方法的近似去噪理论与方法，而无须任何约束条件和先验知识，故适用范围广泛。

射频识别技术(Radio Frequency Identification，RFID)是一种非接触式自动识别技术。该技术凭借标签体积小、成本低、无需接触、多目标同时识别等特征已开始广泛应用于零售业、物流与供应链、图书管理、交通等实时监控和跟踪领域，并成为物联网不可或缺的支撑技术。然而，由于射频技术自身的不稳定性和易受周围环境的影响，来自读写器的数据通常不完整且具有噪声，严重阻碍RFID技术的推广应用。为提高RFID数据流的完整性、可靠性及可用性，本项目着重围绕RFID数据流中存在的漏读、交叉读和重复读等三类典型噪声，深入研究不确定RFID数据流无约束近似去噪理论与技术，主要研究内容包括：不确定RFID数据流模型、统一近似去噪框架、基于频繁情节挖掘的RFID数据清洗策略、基于频繁模式挖掘的RFID数据流清洗策略、移动RFID对象概率清洗策略、冗余RFID数据流近似滤重策略、基于社会网络预测的RFID数据流清洗策略、移动Ad hoc网络检查点策略、RFID对象追溯技术及原型系统开发等，并取得了较多研究成果。本项目的主要特色表现为：围绕RFID数据流的不确定性，提出了一系列基于概率模型的近似去噪策略，并通过实验验证了所提出方法和策略的可行性、正确性和有效性。因此，开展本项目研究既大大丰富了RFID数据流清洗理论，又能为RFID应用推广提供有效技术支撑。