基于层次熵的液压泵特征向量提取方法研究

为了对液压泵特征进行更有效地提取，在样本熵和多尺度熵的基础上提出了基于层次熵（Hierarchical Entropy）的液压泵特征向量提取方法。首先对液压泵振动信号利用层次分解方法进行分解，得到若干节点信号，然后将所有节点信号的样本熵值作为特征向量，结合支持向量机对液压泵进行模式识别。实验数据表明，以样本熵为特征向量的方法四种液压泵状态的样本熵值差别不大，模式识别准确率较低。以多尺度熵为特征向量的方法各状态之间直观区别较明显，模式识别准确率显著提高。以层次熵为特征向量的方法虽然各状态间直观区别不明显，但由于较多尺度熵而言，层次熵不仅考虑了时间序列的“低频”成分，同时考虑了其“高频”成分，更精确和完整地描述了液压泵振动信号的特征，所以模式识别准确率最高。实验数据比较结果验证了该方法的有效性。

Method of Feature Extraction Based on Hierarchical Entropy for Hydraulic Pump

 For purpose of more effective feature extraction of hydraulic pump, a new approach based on HE (hierarchical entropy) is proposed based on sample entropy and multi scale entropy. Firstly, vibration signals of hydraulic pump is decomposed into a number of hierarchical decomposition nodes. Secondly, sample entropy of hierarchical decomposition nodes is calculated as the feature vectors of hydraulic pump. Finally, feature vectors are put into support vector machine to pattern recognition. Analysis of experimental data indicates that, if sample entropy is treated as feature vectors, the difference of sample entropy of four hydraulic pumps′ conditions is not big enough, which result in low accuracy in pattern recognition. If multi scale entropy is treated as feature vectors, the distinction between multi scale entropy of four hydraulic pumps' conditions is large enough, which gives rise to high accuracy in pattern recognition. If HE is treated as feature vectors, the difference of four hydraulic pumps′ conditions is not much enough. Because HE takes both low and high frequency signals into account, it is sufficient and accurate to describe the time series, which leads to a highest accuracy in pattern recognition. The compare of three methods proves the advantage of HE.