快速路网实时OD预测的时间颗粒度选择

Time Granularity Selection for Expressway OD Realtime Prediction

摘要:

常见的基于实测数据的origindestination(OD)预测方法分为两类：一类基于历史信息，即根据上一天或上一周同一日相同时段的数据进行预测，简称同比预测法；另一类则是根据同一天相邻时间段的数据预测本时段的OD，简称环比预测法.预测所用基础数据的时段长度称为时间颗粒度.时间颗粒度的大小对OD预测结果的稳定性、准确性具有重要影响.针对上海快速路网，采用ADF单位根检验和KMeans聚类分析方法，研究时间颗粒度对预测结果的影响，提出了时间颗粒度选择的建议，同比预测方法相比环比预测法更容易得出稳定、合理的预测结果，30~60 min的时间颗粒度预测效果较好.

Abstract:

Two methods, yearonyear prediction method and circularityratio prediction method, are often resorted to in the realtime origindestination(OD) prediction. The former depends on historical information, such as the data about the previous day or week. The latter depends on the data of adjacent period to predict the realtime OD. The size of time granularity, the time period used in the realtime prediction of expressway network has a significant effect on the stability and accuracy of OD prediction results. The Augmented Dickeyfuller Test(ADF) unit root test and Kmeans algorithm are employed to study the effect of prediction results, and the size of time granularity is recommended. Compared with circularityratio prediction method, yearonyear prediction method can lead to the more stable and accurate result and a more satisfied time granularity is between 30 minutes and one hour.