基于K近邻与主成分分析的短时交通流预测  

Short-term Traffic Flow Forecasting Based on K-nearest Neighbors and Principle Component Analysis

作　　者：李翠 黄侃[1] 李霞 LI Cui;HUANG Kan;LI Xia(School of Information engineering,Jiangxi Vocational and Technical College of Communications,Nanchang 330013)

机构地区：[1]江西交通职业技术学院信息工程学院,南昌330013

出　　处：《公路交通技术》2022年第3期138-144,共7页Technology of Highway and Transport

基　　金：江西省交通运输厅科技项目(2020X0012);江西省教育厅科学技术研究项目(191325)。

摘　　要：为有效预测高速公路短时交通流,提出一种将K近邻(KNN)法和主成分分析(PCA)法相结合的KNN-PCA法。工作流程为:1)采用KNN法筛选出与当前状态相似的多条交通流(即近邻);2)将所有相似交通流延拓至待预测时刻,并将其表达为矩阵形式后进行PCA以得到主成分;3)将当前交通流延拓至预测时刻并将其表达为少数起主要贡献主成分的线性组合;4)采用最小二乘法直接得到这些主成分的组合系数估计值;5)基于组合系数估计值,将这些主成分再次线性组合后,直接提取预测时刻的交通量。基于高速公路实测交通流进行数据分析,应用结果表明,KNN-PCA法原理清晰,计算简单,具有良好的预测能力,能很好地预测短时交通流的变化趋势。In order to effectively predict the short-term traffic flow of expressways,a new algorithm named as KNN-PCA is proposed,which combines the K-nearest neighbor(KNN)method and the principle component analysis(PCA)method.The working process is that 1)similar traffic flows which resemble the current one are selected by the KNN method;2)All similar traffic flows are prolonged to the predicted time and expressed in the form of a matrix.The PCA analysis is then conducted for the matrix containing prolonged similar traffic flows;3)The current traffic flow is also prolonged to the predicted time and treated as the linear combination of several principle components which have the greatest contributions;4)Least square method is then utilized to estimate the combinational coefficients of these principle components;5)These principle components are linearly combined again by using the obtained combinational coefficients,with the traffic flow volume at the predicted time extracted directly.The analysis based on field-measured traffic flow data of an expressway shows that,the proposed KNN-PCA algorithm has a clear principle and can be simply computed.It also exhibits a good forecasting performance,with the changing trend of the short-term traffic flow well predicted.

关 键 词：短时交通流 高速公路 最近邻 主成分分析 线性组合 

分 类 号：U491.14[交通运输工程—交通运输规划与管理]