轨迹大数据异常检测:研究进展及系统框架

定位技术与普适计算的蓬勃发展催生了轨迹大数据，轨迹大数据表现为定位设备所产生的大规模高速数据流。及时、有效地对以数据流形式出现的轨迹大数据进行分析处理，可以发现隐含在轨迹数据中的异常现象，从而服务于城市规划、交通管理、安全管控等应用。受限于轨迹大数据固有的不确定性、无限性、时变进化性、稀疏性和偏态分布性等特征，传统的异常检测技术不能直接应用于轨迹大数据的异常检测。由于静态轨迹数据集的异常检测方法通常假定数据分布先验已知，忽视了轨迹数据的时间特征，也不能评测轨迹大数据中动态演化的异常行为。面对轨迹大数据低劣的数据质量和快速的数据更新，需要利用有限的系统资源处理因时变带来的概念漂移，实时检测多样化的轨迹异常，分析轨迹异常间的因果联系，继而识别更大时空区域内进化的、关联的轨迹异常，这是轨迹大数据异常检测的核心研究内容。此外,融合与位置服务应用相关的多源异质数据，剖析异常轨迹的起因以及其隐含的异常事件，也是轨迹大数据异常检测当下亟待研究的问题。为解决上述问题，对轨迹异常检测技术的研究成果进行了分类总结。针对现有轨迹异常检测方法的局限性，提出了轨迹大数据异常检测的系统架构。最后，在面向轨迹流的在线异常检测、轨迹异常的演化分析、轨迹异常检测系统的基准评测、异常检测结果语义分析的数据融合、以及轨迹异常检测的可视化技术等方面探讨了今后的研究工作。

Anomaly Detection for Trajectory Big Data: Advancements and Framework

The thriving development of positioning technology and pervasive computinghave given rise totrajectory big data, which is the high speed trajectory datastream that originated from positioning devices. Analyzing trajectory big data timely and effectively, enables us discover the abnormal patterns that hid in trajectory data streams, and further provide effective support to application fields such as urban planning, traffic management, and security controlling. The traditional anomaly detection algorithms cannot be applied to outlier detection in trajectory big data directly, which is due to the uncertainty, unlimitedness, time-varying evolvability, sparsity and skewness distribution characteristics of trajectories. In addition, most of trajectory outlier detection methods that designed for static trajectory dataset, usually assume a priori known data distribution, and discard the temporal property of trajectory data, and thus are unsuitable for identifying the evolutionary trajectory outlier. When dealing with huge amount of low-quality trajectory big data, we need to consider a series of problems, which include coping with the concept drifts of time-varying data distribution in limited system resources, online detecting trajectory outliers, analyzing causal interactions among traffic outliers, identifying the evolutionary related trajectory outlier in larger spatial- temporal regions, and analyzing the hidden abnormal events and the root cause in trajectory anomalies by using application related multi-source heterogeneous data. Aiming at solving the above-mentioned problems, we review the existing trajectory outlier detecting techniques from several categories, describe the system architecture of outlier detection in trajectory big data, and finally discuss the directions to be studied, which includes outlier detection in trajectory stream, visualized analyzing and evolutionary analysis in trajectory outlier detection, benchmark for trajectory outlier detection system, and data fusion in semantic analyzing for anomaly detection results.