一种调度区段晚点时长的神经网络预测模型

晚点是区段内列车运行受到扰动后出现的时刻表偏移现象,为分析和预测晚点的发生,相关研究通常采用晚点传播分析、实绩数据统计的方法改善模型输出结果。在现有分析方法的基础上,设计了初始晚点和连带晚点的分类方法,将列车在调度区段的开行转化为有向图表示,并通过分析有向弧内的计划时间饱和度,实现了晚点的分类与传播路径的确定。在分类方法提供的数据基础上,提出了晚点预测模型,采用反向传播神经网络预测晚点时长。组合模型使用北京铁路局某调度区段的实际运行数据进行验证,结果表明允许误差为5 min时,神经网络的晚点时长预测准确率为85.5%,网络受突发事件影响较大,模型拟合复杂数据关系的能力需要进一步改善。

A Prediction Model for Timetable Delays in Dispatching Area Using Neural Network

Train delay can be defined as timetable deviation under perturbations in the dispatching area.To analyze and predict delays,relevant researches focus on delay propagation analysis and data statistics to improve model outputs.On the basis of existing analytic methods,this paper puts forward a delay classification method,which expresses train operations in the dispatching area in the form of directed graph.The time saturation rates of all arcs are analyzed to realize the classification of primary and secondary delays and their propagation paths.Based on the data from classification method,a prediction model using BP neural network is recommended to estimate delay length based on counter propagation of neural network.The prediction model is further examined in a dispatching area in Beijing Railway Administration.The results indicate that the accuracy of model prediction is 85.5% with a permissible error of 3%,the network is greatly influenced by contingencies and the model needs further improvements to fit complex data relations.