基于随机森林算法的在线轨道平面线形判别方法

Online Track Plane Alignment Discrimination Method Based on Random Forest Algorithm

为提高现有轨道检测系统轨道平面线形判别的准确性,基于随机森林算法提出一种在线轨道平面线形判别方法。首先利用设备台账和轨道几何检测数据制作轨道平面线形曲线半径识别标准数据库;然后对轨道几何检测数据进行多维度特征提取,运用随机森林算法在样本个数和特征个数上的双重随机构建分类模型并对主要参数进行网格寻优以提高模型分类准确率;最后将模型嵌入到轨道几何参数实时检测处理软件中。该方法离线识别准确率提升至90%以上,能够根据轨道平面曲线半径的分类结果自动切换线形判别关键参数,识别出半径在150~12 000 m的全部轨道平面曲线。

In order to improve the accuracy of track plane alignment discrimination of existing track detection systems,an on-line track plane alignment discrimination method was proposed based on random forest algorithm. Firstly,the standard database of track plane alignment curve radius identification was made by using the standing book and track geometry detection data. Then,the multi-dimensional feature extraction was carried out on the track geometry detection data,the double random classification model was constructed by using the random forest algorithm in the number of samples and the number of features,and the grid optimization of the main parameters was carried out to improve the classification accuracy of the model. Finally,the model was embedded into the real-time detection and processing software of track geometric parameters. The off-line identification accuracy of this method is improved to more than 90%.It can automatically switch the key parameters of linear discrimination according to the classification results of track plane curve radius,and identify all track plane curves with radius of 150~12 000 m.