基于PSO-SVM的居民出行方式预测模型  

Prediction Model of Residents' Trip Mode Based on PSO-SVM

作　　者：许铁[1] 高林杰[2] 景鹏[2] 陈东清[3] 

机构地区：[1]福建交通职业技术学院,福州350007 [2]上海交通大学交通运输工程研究所,上海200052 [3]福州大学管理学院,福州350001

出　　处：《交通运输系统工程与信息》2011年第5期155-161,共7页Journal of Transportation Systems Engineering and Information Technology

基　　金：国家自然科学基金项目(50808123)

摘　　要：居民出行方式选择是一个较为复杂的非线性问题,受到的影响因素众多。提出采用支持向量机方法构建了居民出行方式选择模型,并以交叉验证意义下的分类准确率作为适应度函数,利用粒子群算法对支持向量机参数优化选择,避免参数设定的随机性,减少参数选择的工作量.通过实证研究表明,利用粒子群算法优化支持向量机的参数是可行的,支持向量机方法相对于BP神经网络,对居民出行方式预测有更高的精度.预测精度比BP神经网络提高了将近5个百分点,建模样本和测试样本的分类精度分别达到86.20%和82.31%.所构建的模型可用于居民出行方式预测,这对城市交通规划,出行需求预测具有现实指导意义.Resident trip mode choice is a complex nonlinear problem which is affected by many factors.This paper models the resident trip mode with the support vector machine（SVM） method.It uses the classification accuracy as fitness function in the sense of cross validation and then adopts the particle swarm optimization（PSO） algorithm to select parameters.The choice randomness of specifying the parameters are avoided and the workload of the parameter selection are reduced.The empirical studies indicate that the PSO based parameters optimization in SVM is feasible.Compared with the BP neural network model,the SVM model achieves a better prediction performance for resident trip mode choice.The prediction accuracy of the SVM model is improved by 5% to BP neural network model.The SVM model obtains the forecasting accuracy rate of 86.20% to the model setting and 82.31% to the test samples.The empirical results show that the model can be used for resident trip mode forecasting and provides a practical guidance to urban traffic planning and travel demand forecasting.

关 键 词：城市交通 出行方式预测 支持向量机 粒子群算法 参数选择 

分 类 号：U491.12[交通运输工程—交通运输规划与管理]