基于近邻和SVM的公交停靠时间预测

Prediction of Bus Dwell Time Using Support Vector Machine Based on Near Neighbors

为提高公交行程时间预测的准确性,提升公交系统的整体服务水平,提出了一种公交车站点停靠时间预测模型。在考虑上下车人数、在车人数和天气状况等因素对公交停靠时间的影响下,采用训练集样本预筛选操作,分析样本筛选过程中不同抽样率对预测效果的影响,比较基于不同核函数的支持向量机在预测精度上的差异。选取佛山市301路公交线中有代表性的10个站点,用以检验基于近邻的支持向量机模型的预测效果,并分析不同站点的特性。试验结果表明,所提出的模型可以达到较高的预测精度,决定系数为0.4255,均方根误差为9.4737,且计算时间与不进行预筛选时相比,降低约40%。训练集数据的预筛选过程可以缩短模型的计算时间并且降低预测误差,而基于线性核函数的支持向量机比基于其他核函数的预测效果更好。

To improve the accuracy of predicting bus travel times and the overall service level of a transit system,a bus dwell time prediction model is proposed in this study.Three contributing factors are considered:the number of boarding and alighting passengers,the number of passengers in a bus,and current weather conditions.A training set is selected in advance,and the effects of different selection rates and kernel functions on the prediction performance are analyzed.Ten typical stations on the No.301 bus line in Foshan,China are chosen to test the prediction performance of the support vector machine based on near neighbors method,and the properties of each stop are analyzed.The results indicate that the proposed model achieves high accuracy,namely,an R-square(R2)value of 0.4255 and a root mean square error(RMSE)of 9.4737.The computation time is reduced by approximately 40%as compared with the model without data preselection.The preselection process for the training data set can shorten the calculation time and reduce the prediction error.In addition,the support vector machine based on a linear kernel function performs better than those methods based on other kernel functions.