Abstract: Focusing on the phenomenon that the real traffic databases are highly susceptible to noisy, missing, and inconsistent data, a model is built based on least square Support Vector Machine（SVM） to detect the traffic abnormal data. It transforms the time series data into vector data with the method of phase-space reconstruction. Based on the SVM, it establishes the regression analyse model with the training data before distinguishing abnormal data by calculating the residual error between actual value and predictive value. With the empirical research in Bei Er Duan of the First Circle Road of Chengdu and the comparative analysis with the traditional detection method, the results prove the effectiveness of the proposed detection method.

Key words: traffic flow, discrepant data, support vector machine, least square

摘要： 针对道路交通系统实时交通流数据普遍存在的异常现象，提出一种基于最小二乘支持向量机的交通异常数据检测方法。运用相空间重构技术，将时间序列数据转换为矢量数据。运用训练数据构建基于最小二乘支持向量机的回归估计模型，通过计算实际值与预测值之间的残差来判别异常数据。以成都市一环路北二段进行实证研究，并与传统检测方法比较分析，结果证实该检测方法的有效性。

关键词: 交通流, 异常数据, 支持向量机, 最小二乘法