基于奇异滤波及稳健回归的轨道曲线主点定位方法研究

曲线主点位置的准确获取是线路整正的基础。在缺失线形标记的情况下如何准确地计算出轨道主点的位置一直是业界广泛关注的问题。在分析曲线轨道正矢图形的特征后,提出基于奇异分解和稳健回归的曲线主点定位方法:利用相对测量正矢数据构造截断矩阵进行奇异值分解滤除其中幅值较大的冲击部分,在此基础上通过稳健回归减少数据微小变化带来的拟合偏差,得到主点的具体位置。实验表明:该方法能够基于相对测量的正矢数据推算出曲线各个主点的位置,受分段点选取偏差影响较小,适合工程应用。

Localization Algorithm for Track Curve Feature Point Based on Singular Value Decomposition and Robust Regression

The localization of feature points on the curve is essential in track maintenance. How tocalculate the position of the feature points accurately without line type mark remains the problem ofconcern by the industry. Based on the analysis of the characteristics of curve version diagram, this paperproposes a method based on singular value decomposition algorithm and robust regression to locate trackcurve feature point. The singular value decomposition is used to filter shock component in the vector databy means of interceptive matrixes, and robust regression is employed to reduce fitness bias due to slightdata changes and compute feature point. Test results show that this method can find feature point rapidlyand accurately, less affected by segment point selection bias and is suitable for engineering application.