利用XGBoost和SVR算法的地铁站客流量模型研究

The Research on the Models of Metro Traffic Passenger Flows Based on the XGBoost and SVR Algorithms

为提高城市地铁运营管理能力和及时缓解高峰期大客流压力,对地铁站客流量模型进行研究。首先,利用某大城市地铁站客流量统计大数据以及天气状况(如雨天、雾天、平均温度以及平均湿度等)特征,提出了基于XGBoost(extreme gradient boosting)与支持向量回归机机(support vector regression,SVR)的地铁站客流量分析模型。然后,采用交叉验证的方法进行调参来解决模型训练过程中高偏差或者高方差问题以提高模型性能。最后,对所建立的地铁站客流量模型进行验证。结果表明,梯度提升决策树和支持向量机两种模型的地铁客流量预测准确率分别为82.5%和54.5%,均优于传统的预测模型。

In order to improve the management ability of urban subway operation and timely relief of the stress of large passenger flow during the rush hour, the models of metro traffic passenger flows is studied. First of all, the models of metro traffic passenger flows are are proposed based on Extreme Gradient Boosting(XGBoost) and Support Vector Regression(SVR) taking advantage of statistical big data of metro traffic passenger flows of a big city and characteristics of weather conditions(such as rain, fog, average temperature and average humidity). Then, the cross validation method is used to adjust parameters for solving the problems of high deviation or high variance in the process of training to improve the performance of the model. Finally, the models of metro traffic passenger flows is verified. The result shows that the forecast accuracies of two models based on GBDT and SVR on metro traffic passenger flows are respectively 82.5% and 54.5% which are both superior to the traditional prediction model.