作　　者： ; ; ; ;

机构地区： 华南理工大学经济与贸易学院

出　　处： 《小型微型计算机系统》 2015年第5期1142-1147,共6页

摘　　要： 在智能交通的相关研究中,对城市路段的历史通行速度的分析正引起广泛的关注.然而,由于路段速度数据的采集车辆不能实现对全路网的所有路段的完全覆盖,有部分路段出现了长时间的数据缺失现象,这种在同一路段上连续缺失的数据构成了一个缺失序列.现有的对缺失的速度数据的估计算法主要集中在对缺失点数据的估计,因而不适合应用于本问题.研究对这类缺失序列的估计算法,把该问题建模为序列标注问题,并使用条件随机场求解.研究了对原始速度数据的编、解码策略和与被估计路段相关的路段的选择策略,在实际的交通数据上的实验表明了本文的方法在长序列的速度数据估计问题上的有效性. In recent years ,increasing interests have been shown in analyzing the historical link speed data of urban arteries, related applications including periodicity analysis, flock pattern recognization, route searching, and so on. However, due to the limitation that the number of the data collection cars is far less than that of links in the urban arteries, there exists many links which are often lack of traffic speed data for a long time, thus for this kind of links, the long-time-missing data form missing sequences. For most of the current missing speed data imputation algorithms, they only focus on the estimation of separated missing datum points, thus they are not well suited for the estimation of the missing sequence. In the present work, we focus on the missing sequence estimation algorithm. We model the speed sequence estimation problem as a sequence label one, and we employ the Conditional Random Fields （ CRF ） to solve the problem. We propose both of the encode and decode schemes for the speed-valued data, and we also present an algorithm to select the suitable links to help for both of the CRF training and sequence estimating. Empirical experiments show our algorithm promise.

关 键 词： 条件随机场 缺失数据 序列估计

领　　域： [自动化与计算机技术] [自动化与计算机技术]