【摘要】 为了更准确地预测轨道状态劣化趋势,建立了一种新的轨道状态短期预测集成学习模型。建模时以200 m轨道单元为研究对象,根据轨道交通线路特点考虑多种影响轨道状态劣化程度的异质性因素来确定模型变量。首先分别利用Gamma过程、二项logistic回归和支持向量机三种方法构建TQI预测模型,然后利用Stacking集成学习技术将三个单一模型进行组合,形成新的TQI预测集成模型。采用北京地铁1号线16次TQI检测数据对模型进行了训练和测试,并对比不同模型的分类正确率和AUC值。结果表明,Stacking集成模型有效,能够更加准确地预测TQI变化趋势,同时具有更优的泛化性能。更多还原

【Abstract】 In order to predict the deterioration trend of track state more accurately,a new ensemble learning model of track state short-term prediction was established. The 200 m track unit was taken as the research object in the modeling.According to the characteristics of rail transit lines,a variety of heterogeneous factors affecting the deterioration degree of track state were considered to determine the model variables. Firstly,three methods,which are Gamma process,binomial logistic regression and support vector machine,were used to construct the TQI prediction model. Then,three single models were combined by using Stacking ensemble learning technology to form a new TQI prediction integration model. The model was trained and tested with 16 TQI test data of Beijing Metro Line 1,and the classification accuracy and AUC value of different models were compared. The results show that the Stacking integrated model is effective and can predict the trend of TQI more accurately,and has better generalization performance.更多还原