基于集成学习的城市轨道交通乘客路径选择建模

考虑城市轨道交通路网不同OD间客流构成的异质性,基于数据驱动方法,研究不同OD特性与路径属性组合条件下的乘客路径选择机器学习建模问题,提出将模糊聚类与集成学习相结合的建模方法。运用FCM聚类方法将路网站点划分为若干类别,以模糊聚类结果作为子学习器的划分依据,构建了基于支持向量回归机(SVR)的路径选择子学习器,实现了基于OD类别隶属度对多个子学习器进行组合的路径选择集成预测。该方法既能体现客流特性的差异对路径选择的影响,也解决了难以直接获取乘客属性对建模与预测带来的难题,有效提高了模型准确性。以北京城市轨道交通为案例,对新线开通情况下的路径选择进行预测。

Ensemble Learning Based Modeling of Passenger Route Choice on Urban Rail Transit Network

Considering the influence of heterogeneity of passenger flow among different OD pairs of urban rail transit network on route choice,as well as the data-driven method,the problem of machine learning modeling for passenger route choice under different combinations of OD characteristics and route attributes was studied.A modeling method combined with fuzzy clustering and ensemble machine learning was proposed.The FCM clustering algorithm was used to divide stations into several categories,sub-learners were classified according to the fuzzy clustering result,and then a route choice sub-learner model based on Support Vector Regression(SVR)was constructed.Based on fuzzy membership degree of OD category,multiple sub-learners were integrated to realize route choice prediction.The method can reflect the influence of heterogeneity of passenger flow characteristics on route choice,solve the problem caused by the lack of data of passenger attributes on modeling and prediction,improve the accuracy of the route choice model.A case study was given,in which the method was used to predict the results of passenger route choice under the condition of the opening of new lines.