Outlier filtering of empirical travel time data is essential for traffic analyses. Most of the widely applied outlier filtering algorithms are parametric in nature and based on assumed data distributions. The assumption, however, might not hold under unstable traffic conditions. This paper proposes a nonparametric outlier filtering method based on a robust locally weighted regression scatterplot smoothing model. The proposed method identifies outliers based on a data point’s standard residual in the robust local regression model. This approach fits a regression surface with no constraint on parametric distributions and limited influence from outliers. The proposed outlier filtering algorithm can be applied to various data collection technologies and for real-time applications. The performance of the new outlier filtering algorithm is compared with the moving standard deviation method and other traditional filtering algorithms. The test sites include GPS data of an Interstate highway in Indiana and Bluetooth data of an urban arterial roadway in Texas. It is shown that the proposed filtering algorithm has several advantages over the traditional filtering algorithms.

经验旅行时间数据的异常值过滤对于流量分析至关重要。大多数广泛应用的离群值滤波算法本质上是参数化的，并基于假定的数据分布。但是，该假设可能在不稳定的交通状况下不成立。本文提出了一种基于鲁棒局部加权回归散点图平滑模型的非参数离群滤波方法。所提出的方法基于鲁棒局部回归模型中数据点的标准残差来识别离群值。这种方法适合回归表面，对参数分布没有限制，并且离群值的影响也很有限。提出的离群滤波算法可以应用于各种数据收集技术以及实时应用。将新的离群值滤波算法的性能与移动标准差方法和其他传统滤波算法进行了比较。测试地点包括印第安纳州州际公路的GPS数据和德克萨斯州市区主干道的蓝牙数据。结果表明，与传统的滤波算法相比，该滤波算法具有很多优点。