基于支持向量机方法的轨道交通乘客旅行时间短时预测方法研究

Short-time forecasting of passenger travel time based on SVM method

目前,北京市已建成总长456km的轨道交通网络,41个换乘枢纽,已基本形成网络化运营,轨道交通在北京公共交通系统乃至城市交通系统中发挥越来越大的骨干作用.同时,由于北京城市空间调整,城市交通出行活动中心有随城区扩展外移之势,轨道交通出行旅行时间越来越受到关注.因此,本文研究基于历史数据的轨道交通乘客旅行时间预测方法,引入支持向量机回归模型,并针对预测算法中模型参数的选择问题,提出基于遗传算法的预测模型参数寻优算法,用以提高预测精度.最后选取实际轨道交通运营线路的动态信息进行计算,预测结果误差小于10%,获得了较高的精度.生成的旅行时间预测信息,通过乘客信息系统的发布,能够有效减少旅客换乘走行与等待时间,均衡网络客流分布,提高轨道交通网络运营效率.

With the construction of metro system, currently, Beijing has built a total length of 456 km rail network, 41 transfer hubs, has basically formed a metros network. Rail traffic in Beijing＇s public transport system and even urban transport systems play an increasing backbone role. Meanwhile, due to the Beijing city planning adjustment, urban extent bigger and bigger, the potential rail transit trip travel time is long enough to attention. Therefore, this study is based on historical data rail transit passenger travel time prediction methods, using support vector machine regression model. Focus on the selection of the model parameters in the prediction algorithm, the genetic algorithm based optimization parameter prediction model aim to improve the prediction accuracy. Finally, we select the dynamic information of the actual line of metro system to calculate, obtain a high accuracy of predict less than 10%. To broadcasting the travel time prediction results through the PIS （passenger information system） can effectively reduce passenger transfer time, balance network passenger distribution, improve the operational efficiency of metro network.