基于分形维数和BiLSTM的离心泵空化状态识别方法  

Cavitation recognition method of centrifugal pump based on fractal dimension and BiLSTM

作　　者：邹淑云[1] 刘忠[1] 王文豪 喻哲钦 孙旭辉 ZOU Shuyun;LIU Zhong;WANG Wenhao;YU Zheqin;SUN Xuhui(School of Energy&Power Engineering,Changsha University of Science and Technology,Changsha 410114,China;Beijing Engineering Corporation Limited,Beijing 100024,China)

机构地区：[1]长沙理工大学能源与动力工程学院,长沙410114 [2]中国电建集团北京勘测设计研究院有限公司,北京100024

出　　处：《振动与冲击》2025年第4期305-312,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金项目(52079011);湖南省自然科学基金项目(2023JJ30032)。

摘　　要：针对离心泵空化状态下压力脉动信号的非线性和复杂程度以及浅层机器学习方法在数据深度挖掘上的不足,提出一种基于分形维数和双向长短时记忆神经网络的离心泵空化状态识别方法。通过离心泵空化试验获得不同空化状态压力脉动信号。采用固有时间尺度分解对压力脉动信号进行处理,筛选出有效分量,计算其盒维数和关联维数,构建空化分形特征向量。将空化特征向量导入基于双向长短时记忆神经网络的空化状态识别模型。研究结果表明,有效分量的盒维数及关联维数随空化系数的变化具有明显的规律性,且模型识别的准确率高达92.8%,能够实现离心泵空化状态的识别。Aiming at the nonlinearity and complexity of pressure fluctuation signal of centrifugal pump under cavitation state and the deficiency of shallow machine learning method in data deep mining,a cavitation state recognition method of centrifugal pump based on fractal dimension and Bidirectional Long Short⁃Term Memory(BiLSTM)was proposed.The pressure fluctuation signals under different cavitation states were obtained through the cavitation test of centrifugal pump.The Intrinsic Time⁃scale Decomposition(ITD)was used to process the pressure fluctuation signal,the effective components were selected,the box and correlation dimension were calculated,and the cavitation fractal feature vector was constructed.The cavitation feature vector was introduced into the cavitation state recognition model based on BiLSTM.The research results show that the box and correlation dimension of the effective component have obvious regularity with the change of cavitation coefficient,and the accuracy of model recognition is as high as 92.8%,which can realize the identification of cavitation state of centrifugal pump.

关 键 词：离心泵 空化 压力脉动 固有时间尺度分解 分形维数 双向长短时记忆神经网络 

分 类 号：TH311[机械工程—机械制造及自动化]