聚类分析在消除轮轨力信号基线漂移中的应用

Application of cluster analysis in removing baseline wandering of wheel-rail force signal

摘要:

针对轮轨力信号预处理过程中基点难于提取和确认的问题，根据轮轨信号幅值变化的特性，提出了基于分段数据高阶统计量聚类分析的方法来筛选出能够提取基点的数据段。轮轨力信号经过适当分段后，计算数据段的高阶统计量，按数据段的方差和峭度进行基于OPTICS的聚类分析，选取聚类结果靠近零点的分类所对应的数据段的中值作为基点，经曲线拟合基点后即可得到信号的基线漂移干扰。经过仿真数据和实测数据的验证和与其他现有常用方法的对比分析，结果表明该方法在均方误差和信噪比上都优于其他方法。均方误差最高仅为其他方法的0.47%，信噪比至少比其他方法高出23 dB。

Abstract:

Aiming at the problem that it is difficult to extract and confirm the base point in the preprocessing process of wheel rail force signal, according to the variation characteristics of wheel rail signal amplitude, a method based on cluster analysis of higher-order statistics of segmented data is proposed to screen the data segments that can extract the base point. After the signal is properly segmented, the high-order statistics of the data segment are calculated, and the clustering analysis based on OPTICS is carried out according to the variance and kurtosis of the data segment. The median value of the data segment that corresponding to the classification of the clustering result which close to the zero point as the base point, and the baseline wandering interference of the signal can be obtained after curve fitting the base point. The results of verification with simulation data and measured data and comparative analysis with other existing common methods, show that this method is superior to other methods in mean square error and signal-to-noise ratio. The maximum mean square error is only 0.47% of that of other methods, and the signal-to-noise ratio is at least 23 dB higher than that of other methods.