基于R-FPOP变点检测的城市路段旅行时间预测

Prediction of Urban Road Section Travel Time Based on R-FPOP Change Point Detection

针对城市路段旅行时间精准推送的不足,提出一种基于动态规划变点检测算法的旅行时间预测方法。以车牌识别数据为研究对象,利用R-FPOP算法对旅行时间均值变点进行在线检测,研究变点时域分布特征;基于均值变点检测结果,预测旅行时间并给出其预测区间。结果表明:在线检测出的变点能够有效辨识旅行时间的均值突变,变点时域分布主要集中在高峰期;旅行时间预测值对实际序列变化趋势估计准确,推送的预测区间平均覆盖率为79.54%,具有较优的预测精度。论文方法兼顾旅行时间均值突变且建模简单,可为路段旅行时间的在线智能推送及交通需求者的路线规划提供技术支持。

For the shortcomings of the precise pushing of urban road section travel time, a travel time prediction method based on dynamic programming algorithm for change point detection was proposed. The license plate recognition data was used as the research object, and the robust functional pruning optimal partitioning (R -FPOP) algorithm was used to detect the mean change points of travel time sequences online. Further, the temporal distribution characteristics of the change points were studied. Based on the detection results of the mean change points, the road section travel time sequences were predicted and their prediction intervals were given. The results show that the mean change of the travel time sequences can be effectively identified by the detected change points online, and the temporal distributions of the change points were mainly concentrated in the peak period. The predicted values of the travel time were estimated accurately for the change trend of actual sequences. The average coverage ratio of the pushed prediction intervals is 79.54%, and the prediction accuracy of the travel time is excellent. The method takes into account the mean change of the travel time sequences and is simple to model, which could provide technical support for the intelligent pushing of road section travel time online and route planning for traffic demanders.