摘要
非参数回归在交通流预测中已得到广泛应用,但实际使用中存在实时性差的缺陷。为提高非参数回归预测速度,提出了使用空间索引结构R树作为模式库的存储结构,并依据R树空间聚类的特点进行K近邻搜索,最后根据搜索到的近邻点估算未来的交通流量。实验结果表明R树结构下的K近邻搜索速度比线性结构下的搜索速度提高了59.6%,但预测精度下降了8.8%。而通过缩小K近邻搜索中的距离上限这一参数,可以提高预测精度。结果表明当2种结构下的距离上限相同且小于0.02时,R树结构下的预测精度平均高于线性结构下11.9%,且搜索速度也平均提高了30.8%。因此,该算法能够在满足预测精度的条件下有效地提高预测速度,为实时短时交通流预测系统提供了1种实现算法。
Nonparametric regression has been widely employed in traffic flow prediction; however, it cannot meet the requirement of real-time applications. This paper proposes an algorithm to improve the speed of nonparametric regression prediction. The algorithm uses spatial indexing structure of R-tree as the storage structure of the historical database, and modifies K-Nearest neighbor search according to the characteristics of R-tree spatial clustering to estimate the future traffic flow. Numerical experiments demonstrate that K-Nearest neighbor search based on R-tree can improve the speed of prediction by 59.6%, but reduce the accuracy by 8.8% when compared with the linear structure approach. By reducing the distance ceiling, the accuracy can be improved. The results show that when both kinds of structures have the same distance ceiling less than 0.02, K-Nearest neighbor search using R-tree can generally increase the accuracy by 11.9% and speed by 30.8% over the linear structure-based search. Therefore, the algorithm can enhance the forecasting efficiency effectively within the preset accuracy range, representing a better way for real-time traffic flow forecasting.
出处
《交通信息与安全》
2014年第6期65-71,共7页
Journal of Transport Information and Safety
关键词
智能交通系统
交通流预测
非参数回归
R树
K近邻搜索
intelligent transport systems
traffic flow prediction
nonparametric regression
R-tree
K-nearest neighbor search