基于二元特征的断面交通数据异常检测与修正

Anomaly detection and correction of sectional traffic data based on bivariate features

为了确保交通数据的准确性以及为后续研究提供数据质量保障。提取断面交通数据的两元特征向量速度与流量,采用多元高斯模型拟合两元特征概率密度分布,并利用matlab训练模型,对测试数据集进行异常检测;针对检测出的异常数据点,引入小时窗口,建立修正数据集。随后在各修正窗口内采用波动性处理优化的灰色马尔可夫模型,对上述异常点进行修正。以汕昆高速K2077断面为例,对其采集数据进行模拟计算及分析,结果表明:多元高斯模型对断面交通数据的二元特征拟合效果良好,异常检测作用突出;引入修正窗口简化了数据修正的运算过程,避免了大量历史数据冗余使用;波动性处理优化弥补了灰色马尔可夫模型针对随机性数据性能下降的缺陷,大幅提升了数据修正的准确率。

In order to ensure the accuracy of traffic data and provide data quality assurance for the follow-up research,the bivariate eigenvector velocity and flow of sectional traffic data were extracted.The probability density distribution of bivariate features was fitted by using the multivariate Gauss model,and the anomaly detection of the testing data sets was carried out by using the MATLAB training model.Aiming at the detected abnormal data points,the hourly window was introduced to establish the corrected data set.Then the grey Markov model optimized by volatility processing was applied in each correction window to correct the above abnormal points.Taking the K2077 section of Shan-Kun expressway as an example,the collected data were simulated and analyzed.The results showed that the fitting effect of multivariate Gauss model on the bivariate features of sectional traffic data was good,with a prominent role in anomaly detection.The introduction of correction window simplified the operation process of data correction,and avoided the redundant use of large amount of historical data.The optimization of volatility processing made up for the shortcoming of grey Markov model in the performance reduction of random data,and greatly improved the accuracy of data correction.