城轨列车在车站停车误差估计模型与在线学习算法的研究  

Research on the Error Estimation Models and Online Learning Algorithms for Train Station Parking in Urban Rail Transit

作　　者：陈德旺[1] 唐涛[1] 郜春海[1] 穆瑞琦[1] 

机构地区：[1]北京交通大学轨道交通控制与安全国家重点实验室,北京100044

出　　处：《中国铁道科学》2010年第6期122-127,共6页China Railway Science

基　　金：国家自然科学基金资助重点项目(60736047;60634010);轨道交通控制与安全国家重点实验室(北京交通大学)自主课题(RCS2008ZZ001);轨道交通控制与安全国家重点实验室(北京交通大学)开放基金资助项目(RCS2008K007)

摘　　要：根据城轨列车制动特性,对列车运动方程做2项简化。一是忽略空气阻力和坡度的影响;二是分别假设制动力传递有延时及减速度不变或减速度是初始速度的线性函数。由此推导出2个简化的列车停车误差估计模型和模型参数之间的线性关系函数式,并给出模型参数在线学习算法,以克服停车过程中的各种非线性因素的影响,提高停车精度。根据统计学原理,采用5个评价指标对模型的性能进行评价,采用停车误差估计判断停车精度是否满足停车可靠性的要求。利用实测停车数据对模型和在线学习算法进行验证和比较。结果表明:提出的简化模型和在线学习算法,能有效降低停车误差,并纠正误差分布的有偏性;停车误差在大于99.5%的情况下满足30cm停车精度的可靠性要求;模型1的效果比模型2略好。Based on the braking characteristics of train station parking in urban rail transit,two simplifications about the train dynamics equation were made：one was neglecting the influence caused by the air resistance and the gradient,the other was assuming that there was a braking delay and a constant deceleration or the deceleration was a linear function of the initial speed.Then,two simplified models were developed to estimate the train station parking error.Furthermore,the linear relationships between the parameters of the two models were induced and the online learning algorithm for updating the models＇parameters was developed to overcome the influence by the nonlinear factors in parking to increase the parking accuracy.According to the statistics principle,five evaluation indices were adopted to evaluate the performance of the models and the parking error estimation was adopted to judge whether the parking error could meet the requirements of parking reliability.The field train parking data were used to validate and compare the performance of the models and the online learning algorithm.The results show that the proposed simplified models and the online learning algorithms are effective in reducing the parking error and correct the bias of the error distribution.When the parking error is greater than 99.5%,it can meet the 30cm reliabil-ity requirements for the parking accuracy.Model 1achieves better performance than model 2.

关 键 词：城市轨道交通 车站停车 停车精度 误差估计模型 在线学习算法 

分 类 号：U231.6[交通运输工程—道路与铁道工程]