小幅蛇行异常是剧烈蛇行失稳的征兆，它不仅影响乘坐舒适性，导致轮轨疲劳接触，而且随着轮轨磨损加剧、列车服役时间增长、运行速度提高，小幅蛇行会不断加剧，特别是在抗蛇行减震器失效的状况下，可能会引起列车脱轨，严重影响行车安全。但现有的高速列车转向架峰值监测法不能监测小幅蛇行异常。针对该问题，提出一种基于经验模态分解和流形学习的特征提取方法。首先，利用EMD分解得到多个固有模态函数（IMF），计算每个IMF的样本熵，作为初步提取特征；然后利用流形学习方法对初步提取的特征进一步提取；最后利用最小二乘法支持向量机对特征提取方法进行评估，并将该方法应用于高速列车320~350 km/h状态下小幅蛇行异常识别中，小幅蛇行异常的识别率达到100%。结果证明：EMD-ISOMAP方法能够有效识别小幅蛇行异常，识别效果优于基于小波变换特征提取方法；该方法降低特征数据复杂度的同时，还增强状态识别的分类性能。%Small hunting is a sign of severe hunting instability. Small hunting hinders riding experience and leads to the fatigue of wheel / rail contact,besides,it aggravates constantly along with the aggravating of the wear of the wheel / rail and the increasing of service time and speed of high-speed train. Especially under the condition of anti-hunting damper failure, small hunting may cause train derailment, putting serious threat to safety. But the existing bogie lateral acceleration peak value monitoring method fails to monitor small hunting. A feature extraction method based on EMD and manifold learning is proposed. Firstly,original signals are decomposed to a finite number of intrinsic mode functions by using EMD. The sample entropy of each IMF is calculated as the preliminary feature. Secondly, the preliminary feature is further extracted by using manifold learning. Finally,the least squares support vector machine is employed to evaluate the feature extraction method. Moreover, the proposed method was applied to the recognition of small hunting of high-speed train running at the speed of 320-350 km/h:the recognition rate of small hunting anomaly is 100%. The result shows that EMD-ISOMAP method can identify small hunting effectively, and the result has better recognition effect than the method based on the sample entropy of wavelet transform. The EMD-ISOMAP method reduces the complexity of the feature data,while also enhances the classification performance of state recognition.
展开▼
