基于神经网络的交通发生量预测研究  

Research on trip generation forecasting based on BP neural network

作　　者：孙健[1,2] 陈书恺 竺寒冰 

机构地区：[1]上海交通大学船舶海洋与建筑工程学院海洋工程国家重点实验室,上海200240 [2]上海交通大学船舶海洋与建筑工程学院,上海200240 [3]大连海事大学交通运输管理学院,辽宁大连116026

出　　处：《西安建筑科技大学学报（自然科学版）》2015年第2期204-209,共6页Journal of Xi'an University of Architecture & Technology(Natural Science Edition)

基　　金：国家自然科学基金项目(71101109);北京大学-美国林肯土地政策研究院论文资助项目(DS20140901);长沙理工大学公路工程教育部重点实验室开放基金项目(kfj120108)

摘　　要：交通发生吸引量预测是交通规划四阶段的首要步骤,其预测结果是城市规划布局及交通设施建设发展的重要依据.为了提高交通发生量预测准确性,利用K-means聚类分析对交通小区进行分组;对同组内样本小区各项土地利用及人口就业指标进行主成分分析,通过计算主成分载荷率为选择预测影响因素提供依据;针对各组样本分别建立BP神经网络模型,以土地利用和人口数据作为输入变量,小区交通发生量作为输出变量,以大连市城市交通调查数据为例对上述方法进行检验,并与传统回归模型预测结果进行比较.结果表明,在数据预处理基础上建立的BP神经网络模型具有较高预测精度.Forecasting trip generation and attraction is the first component of the four-stage method in transportation planning, which determines the urban layout and construction of traffic facilities. To improve the accuracy of trip generation forecasting, K-means cluster analysis was used to divide traffic zones into several groups according to the population and employment. Principal component analysis was conducted to calculate the loading rate to principal components, providing the basis for choosing the influence factor. Finally, BP （Back Propagation） neural networks were set up to forecast trip generation; the input included land-use and population of each traffic zone; and the output was the trip generation. The methods were testified with the traffic survey data from city of Dalian, Liaoning province. Moreover, the results were compared with those obtained from multiple regression model. It is indicated that the BP neural network based on data pre-process produces better results in trip generation forecasting.

关 键 词：预测方法 BP神经网络 交通发生量 聚类分析 主成分分析 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]