快速路行程速度突变段挖掘和基于相似匹配的短时预测算法  

VELOCITY FLUCTUATION PERIOD MINING FOR EXPRESSWAY TRAFFIC FLOW AND SHORT-TERM FORECASTING ALGORITHM BASED ON SIMILARITY MATCHING

作　　者：李心玥 杨卫东[1] 张徵[1] 许海波[1] Li Xinyue Yang Weidong Zhang Zheng Xu Haibo(School of Computer Science, Fudan University, Shanghai 201203, China)

机构地区：[1]复旦大学计算机科学技术学院,上海201203

出　　处：《计算机应用与软件》2017年第3期38-44,86,共8页Computer Applications and Software

基　　金：国家十二五专项(CHINARE2012-04-07;MJ-Y-2011-39);上海市重点基础研究项目(08JC402500);上海市高新技术创新专项(11CH-03);海洋公益项目(201405031-04);上海市科学技术委员会科研计划项目(14511107403)

摘　　要：城市快速路交通流具有明显的畅通和拥塞时段交错的特征,其间行程速度产生较大变化。基于上海快速路线圈感应器采集的数据,首先提出一种在交通时间序列上线性时间挖掘行程速度突变段的滑动窗口方法,解决了识别拥塞起止时刻等需求。然后,构建突变段历史样本数据库和自定义索引,提出一套经过多重优化的基于相似度匹配的预测模型,达到对行程速度短时预测的目的,相比传统的回归方法更简单实用。最后,利用大量实际数据对两套模型的效果和性能进行了检验。结果表明,挖掘算法通过简单的参数调校可完成不同尺度的突变段查询,而预测算法能有效满足实时查询的性能要求,15 min的预测精度能达到90%左右。Traffic flow on urban expressways has obvious characteristics that its travel velocity changes drastically between the smooth condition and the congestion. On the basis of data collected by coil sensors on Shanghai expressway, we propose a promising linear sliding window method to mine the travel velocity fluctuation periods in traffic time series, solving the requirements such as identifying the start and stop moment of congestion. Then, we construct the history sample database of fluctuation periods and the custom index, and propose an optimized forecasting model based on similarity matching to achieve the purpose of short-term forecasting of travel velocity. The model is more practical and intuitive than traditional regression model. Finally, we use huge amounts of real data to test the effectiveness and performance of these two models. Experimental results show that the mining algorithm can complete fluctuation period query at different scales through simple parameter tuning, while the forecasting model can effectively meet the performance requirement of real-time query and reach a high forecasting accuracy around 90% in 15 minutes.

关 键 词：时间序列 行程速度 突变段 滑动窗口 短时交通预测 相似度 索引构造 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]