关键词: 旅行时间, 噪音数据, 混合模型, 对数正态分布, 数据聚类

Abstract:

Travel time samples collected from a special road for a time period can be obtained from the Automatic Number Plate Recognition（ANPR） data. However, the samples often contain a fair amount of noise travel time that can not represent the usual traffic conditions. Filtering out these noise travel time can derive the valuable valid travel time representing the usual traffic conditions. Hence, an algorithm is presented to fit the distribution of data samples with the log-normal mixture model, and the two criteria of optimal [K] value of component distribution are given to make a best classification of two types of travel time with respect to the characteristics of right tail portion of distribution of noise travel time, so the noise travel time can be identified and the valid travel time can be extracted. The algorithm also can discern the samples with the weak characteristics of noise travel time, and extract the valid travel time from the samples between the 10th and 90th percentile. The experiments using the algorithm on the actual ANPR data coming from the buses and non-bus vehicles gain a good effect in identifying the noise travel time. The experimental results show that there are obvious differences of mean and standard deviation of travel time for non-bus vehicles and standard deviation for buses before and after filtering the noise travel time. This indicates that the wrong judgment for the running state for the two traffic modes under the normal traffic conditions would be made without filtering out the noise travel time.

Key words: travel time, noise travel time, mixture model, log-normal distribution, data clustering