基于改进二叉树支持向量机的离心泵工况识别方法

Identification method for centrifugal pump working condition based on improved binary tree support vector machine

为保证离心泵的安全高效运行,需要对离心泵的运行工况进行识别研究.首先,使用测试函数对比研究了经验模态分解、集合经验模态分解和互补集合经验模态分解3种振动信号特征提取方法,基于性能最优的特征提取方法提取不同工况下运行的离心泵振动信号特征数据.然后,对支持向量机模型进行改进,提出了一种使用k-means聚类算法优化的二叉树支持向量机模型,并将改进模型应用到离心泵4种不同运行工况的识别中.同时,使用其他2种多分类支持向量机模型作为对比.研究结果表明:3种特种提取方法中,互补集合经验模态分解无模态混叠迹象性,噪声干扰小,性能表现更好;改进支持二叉树向量机模型分类准确率可达82.17%,对设计的4种工况具有很好的分类效果;改进支持二叉树向量机模型结构简单,训练时间短,实时性好,综合性能优于其他2种模型.

In order to ensure the safe and efficient operation of centrifugal pumps,it is necessary to identify the operating conditions of centrifugal pumps.Firstly,three feature extraction methods of vibration signals,empirical mode decomposition,ensemble empirical mode decomposition and complementary ensemble empirical mode decomposition,were compared and studied by using test functions.The feature data of vibration signals of centrifugal pumps under different operating conditions were extracted based on the feature extraction method with optimal performance.Then,the support vector machine model was improved,and a binary tree support vector machine model optimized by k-means clustering algorithm was proposed.The improved model was applied to the identification of four different operating conditions of centrifugal pumps.At the same time,the other two multi-classification support vector machine models were used as comparison.The results show that among the three special extraction me-thods,the complementary ensemble empirical mode decomposition has no modal aliasing sign,less noise interference and better performance.The classification accuracy of the improved support binary tree vector machine model can reach 82.17%,which has a good classification effect on the four working conditions designed.The improved support binary tree vector machine model has simple structure,short training time,good real-time performance and better comprehensive performance than the other two models.