基于时空行为规律挖掘的公交乘客分类方法

应用智能公交系统(APTS)提取个体乘客出行信息，构造了公交出行链，研究了基于时空行为规律挖掘(STBRM)的公交乘客分类方法；应用时间序列表征乘客出行时间特征，利用互相关距离(CCD)算法计算了个体乘客出行时间规律；应用带噪声基于密度的空间聚类(DBSCAN)算法，挖掘了个体乘客的出行空间规律；依据出行强度和出行时空规律，将乘客划分为极少出行、时间规律、空间规律、时空规律和不规律等5个群体；以出行天数、类似上车时间数量和类似上车站点数量为聚类指标，应用K-Means++算法将乘客划分为高规律、中规律和低规律3类，比较了本文提出的STBRM方法和K-Means++聚类方法的分类结果，揭示了2种方法分类结果之间的关系。研究结果表明: 当时段划分长度取5 min，时间规律性判断阈值取3.0时，利用CCD算法识别时间模式规律乘客的效果最佳，与常用的DBSCAN算法相比，识别率提升了14.64%；增加时间窗长度能够提高时间、空间模式规律判定结果的稳定性；时间窗长度达到3周后，空间模式规律的乘客比例下降趋缓，达到6周后趋于稳定；时间窗长度达到2周后，时间模式规律的乘客比例增长趋缓，达到4周后趋于稳定；时间规律、空间规律和时空规律等3类乘客数量仅占总乘客数量的30.4%，但其出行量占到了总出行量的84.7%，公交依赖度很高，应作为公交机构重点保障的对象；本文提出的STBRM方法与K-Means++聚类方法的分类结果具有较强的关联性，规律性极高或极低的群体高度重合。 

Bus passenger classification method based on spatial and temporal behavior regularity mining

Using the advanced public transportation system (APTS) to extract individual passenger travel information, the bus trip-chain was constructed, and the method of bus passenger classification based on the spatial and temporal behavior regularity mining (STBRM) was examined. Time series were used to characterize the travel temporal characteristics of passengers, and the cross-correlation distance (CCD) algorithm was used to calculate the temporal regularity of individual passengers. The density-based spatial clustering of applications with noise (DBSCAN) algorithm was used to mine the travel spatial regularity of individual passengers. According to the travel intensity and spatial-temporal regularity, bus passengers were divided into five groups, including rare travel, regular time, regular space, regular space-time, and irregular. Taking the numbers of travel days, similar boarding times, and similar boarding bus stops as the clustering indexes, the K-means++ algorithm was applied to classify passengers into three categories, namely high regularity, medium regularity, and low regularity. The classification result of the proposed method was compared with the K-means++ clustering method, and the relationship between the two methods was revealed. Research results show that when the time division length is 5 min and the temporal regularity judgment threshold is 3.0, the CCD algorithm has the best identification effect of passengers with temporal shifted patterns. Compared with the DBSCAN algorithm, the recognition rate improves by 14.64%. Increasing the time window length can improve the stability of travel spatial and temporal regularity judgment. When the time window length reaches three weeks, the proportion of passengers in a spatial pattern decreases slowly and becomes stable after six weeks. When the time window length reaches two weeks, the proportion of passengers in a temporal pattern increases slowly and becomes stable after four weeks. The number of passengers for regular time, regular space, and regular space-time accounts for only 30.4% of the total number of passengers, but their number of trips accounts for 84.7% of the total number of trips, therefore, the bus dependence is very high, which should be taken as the key service object of public transport institutions. The classification results of the proposed method and K-means++ clustering method have a strong correlation, and the groups with very high or very low regularity for the two methods have a high degree of overlap. 13 tabs, 9 figs, 31 refs.