Abstract: Based on the large-scale survey data of residents travel, the travel mode choice behavior of travelers in Xining, a typical corridor valley pattern city of loess plateau was explored. Firstly, the socio-economic attributes and travel characteristics of individuals and families were extracted, and the choice of two commuting modes which meant private transportation mode including car travel and taxi travel and public transportation mode was taken as the target variable. Through the significance test, eight decision variables affecting the choice of travel mode were obtained. Then, for the valid samples of 29,960 travel records, the training sample set and test sample set were reasonably divided. Based on this, support vector machine (SVM) and traditional binomial logistic model (BL) were constructed respectively to identify the travel mode choice of commuters. The classification prediction accuracy, overall classification prediction accuracy and average absolute percentage error of three quantitative indicators were selected to compare the classification results of different models. The results show that compared with BL model, SVM has better fitting effect on classified data and good applicability for prediction of travel mode selection. Specifically, the prediction accuracy of SVM is 2.76% higher than that of the BL model for private transportation modes, and 8.08% higher for public transportation. The overall classification accuracy of SVM is 4.82% higher than that of BL model.

Key words: traffic engineering; corridor valley pattern city of loess plateau; Xining city; support vector machine (SVM); travel mode choice

摘要： 为探究典型高原川道型城市（西宁市）的出行者对交通出行方式选择的行为，以大规模的居民出行调查数据为基础，提取个人、家庭社会经济属性及出行特征指标，将包含小汽车出行和出租车出行的私人交通方式与公共交通方式这2项通勤方式选择作为目标变量，并通过显著性检验得出影响出行方式选择的8项决策变量。针对29 960次有效出行样本，合理划分出训练样本集和测试样本集。基于此，分别构建支持向量机(SVM)和传统的二项Binary logistic (BL)模型以识别通勤者的出行方式选择，选取3项定量指标分方式的分类预测精度、总体分类预测精度和平均绝对百分比误差以对比不同模型的分类结果。研究结果表明：相比BL模型，SVM对分类数据具有更好的拟合效果，对出行方式选择的预测适用性良好，具体来说，对私人交通方式的预测，SVM的预测准确率比BL模型的预测准确率高出8.08%，公共交通则高出了2.76%；SVM的总体分类准确率比BL模型的预测准确率高出4.82%。

关键词: 交通工程；高原川道型城市；西宁市；支持向量机（SVM）；出行方式选择