基于GBDT的轨道不平顺状态评价模型研究

基于轨道几何动态检测数据和车载式线路检查仪(晃车仪)数据,通过随机森林模型分析轨道几何特征与水平、垂直晃车相关性,并结合车辆动态响应利用迭代决策树(Gradient Boosting Decision Tree,GBDT)算法建立轨道不平顺状态评价模型,利用该模型对一客运专线实测轨道几何数据和晃车仪数据进行数据训练和预测。结果表明,模型能够识别超出现有幅值评判标准对车辆运行有显著影响的轨道病害区段,有益于完善轨道几何不平顺评价体系及工务设备养护维修。

Research on Evaluation Model of Track Irregularity Based on Gradient Boosting Decision Tree

Based on the dynamic detection data of track geometry and the data of vehicle mounted track detector(vehicle shaking instrument),the correlation between track geometric characteristics and horizontal and vertical vehicleshaking was analyzed by random forest model.Combined with the dynamic response of vehicles,the evaluation model oftrack irregularity was established by using gradient boosting decision tree(GBDT).Based on the measured trackgeometry data and vehicle shaking instrument data of a passenger dedicated line,the evaluation model was used for datatraining and prediction.The results show that the model can identify the track disease section which has significantimpact on vehicle operation beyond the existing amplitude evaluation standard,and has an important guiding significancefor improving the evaluation system of track geometry irregularity and maintenance and repair of track equipment.