基于朴素贝叶斯分类的居民出行起讫点识别方法

针对手机信令数据存在的精度不高、时间间隔大、信号"乒乓切换"等问题，提出一种基于朴素贝叶斯分类（NBC）的方法来利用手机定位数据识别居民出行起讫点（OD）。首先，利用80位志愿者连续1个月记录的出行活动数据，依据职住距离分类统计移动和停留状态下的条件概率分布；其次，建立用于表征用户移动停留状态的两个特征参数指标：方向夹角和最小覆盖圆直径；最后，依据NBC原理计算用户的移动或停留状态概率，将连续两个以上为移动状态的过程集聚为出行OD。利用厦门市移动的手机定位数据的分析结果表明：所提方法得到的人均出行次数的平均绝对百分比误差（MAPE）误差为7.79%，具备较高的精度，出行OD的分析结果可以较好地反映真实出行规律。

Residents' travel origin and destination identification method based on naive Bayes classification

Mobile signaling data has the characteristics of low accuracy, large time interval and the existence of signal "ping-pong switching". In order to identify residents' travel Origin and Destination (OD) using mobile location data, a method based on Naive Bayesian Classification (NBC) was proposed. Firstly, according to the distance between places of residence and working, the travel log data measured by 80 volunteers for one month were classified statistically, and the conditional probability distribution of moving and staying states was obtained. Then, the feature parameters used to represent the user's states of moving and staying were established, including angular separation and minimum covering circle diameter. Finally, the conditional probability distribution of moving and staying states was calculated according to NBC theory, the processes with more than two consecutive moving states were clustered into travel OD. The analysis results on Xiamen mobile location data indicate that the travel time per capita obtained by proposed method has the Mean Absolute Percentage Error (MAPE) of 7.79%, which has a high precision, and the analysis results of travel OD can better reflect real travel rules.