Abstract：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.
陈代兵,袁寿其*,裴吉,王文杰. 基于改进二叉树支持向量机的离心泵工况识别方法[J]. 排灌机械工程学报, 2023, 41(1): 8-15.
CHEN Daibing,YUAN Shouqi*,PEI Ji,WANG Wenjie. Identification method for centrifugal pump working condition based on improved binary tree support vector machine. Journal of Drainage and Irrigation Machinery Engin, 2023, 41(1): 8-15.
