基于迁移学习和CNN-LSTM的水轮机空化状态识别方法  

Cavitation State Recognition Method of Hydraulic TurbineBased on Transfer Learning and CNN-LSTM

作　　者：刘忠[1] 周泽华 邹淑云[1] 刘圳 乔帅程 LIU Zhong;ZHOU Zehua;ZOU Shuyun;LIU Zhen;QIAO Shuaicheng(School of Energy and Power Engineering,Changsha University of Science and Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学能源与动力工程学院,长沙410114

出　　处：《动力工程学报》2024年第10期1533-1540,共8页Journal of Chinese Society of Power Engineering

基　　金：国家自然科学基金资助项目(52079011);湖南省自然科学基金资助项目(2023JJ30032);湖南省研究生科研创新资助项目(CX20220927)。

摘　　要：针对水轮机空化声发射信号中包含较多噪声、依赖人工降噪与特征提取以及深度学习模型准确率极度依赖海量训练数据的问题,提出一种基于迁移学习和卷积神经网络-长短时记忆网络(CNN-LSTM)的水轮机空化状态识别方法。首先,将数据输入CNN中提取隐含特征;然后,在LSTM中提取特征隐含的时序信息并输出空化类型,通过训练网络参数建立基于CNN-LSTM的空化状态识别模型;最后,引入迁移学习对类似工况进行空化状态识别。结果表明:该模型能准确识别出3种不同的水轮机空化类型,其平均识别准确率达到较高水平;与传统深度学习模型相比,该模型在极少样本学习任务中的识别准确率具有明显优势。Aimed at such problems as containing noises in hydraulic turbine cavitation acoustic emission signals,noise reduction and feature extraction relying on manual operation,and the accuracy of deep learning model heavily depending on massive training data,a method of hydraulic turbine cavitation state recognition was proposed based on transfer learning and convolutional neural networks-long short term memory network(CNN-LSTM).First,the data was input into CNN to extract the hidden features.Then,the temporal information implied in the feature was extracted from LSTM and the cavitation type was output.The cavitation state recognition model based on CNN-LSTM was established by training network parameters.Finally,cavitation state under similar working conditions was identified by introducing transfer learning.Results show that the proposed model can accurately identify three different types of turbine cavitation with average recognition accuracy of high level.Compared with traditional deep learning model,the proposed model has obvious advantages in recognition accuracy in very few sample learning tasks.

关 键 词：水轮机空化 声发射信号 卷积神经网络 迁移学习 长短期记忆网络 

分 类 号：TK73[交通运输工程—轮机工程]