哈尔滨工程大学学报2026，Vol.47Issue(2)：248-254,7.DOI:10.11990/jheu.202405017
基于递归图和卷积神经网络-门控循环单元的水轮机空化状态识别方法
Cavitation state identification of hydraulic turbine based on recurrence plot and CNN-GRU
摘要
Abstract
Aiming at the problem that it is difficult to effectively extract the characteristics of acoustic emission(AE)signals induced by hydraulic turbine cavitation in complex noise interference environments,which affects the accuracy of cavitation state recognition,a water turbine cavitation state recognition method based on recurrence plot and a convo-lutional neural network-gated recurrent unit(RP-CNN-GRU)combined network is proposed.Phase space reconstruction is carried out on the hydraulic turbine cavitation AE signal,recurrence plots under different cavitation states are ob-tained through recursive analysis,and are input as cavitation feature images into the CNN;cavitation characteristics hid-den in the recurrence plot are extracted through the CNN;timing information in the hidden features is extracted in the GRU and cavitation state recognition is completed.Research shows that the cavitation recognition accuracy of the CNN-GRU model taking the recurrence plot dataset as the input is 96.8%,which is higher than other image datasets such as time-frequency graphs and Markov transition field.The average F1 score of this method for identifying cavitation states of hydraulic turbines under multiple operating conditions is 0.94,which has higher recognition accuracy and generaliza-tion performance for feature extraction and classification of nonlinear signals.关键词
水轮机/空化/声发射信号/特征提取/递归图/卷积神经网络/门控循环单元/深度学习Key words
hydraulic turbine/cavitation/acoustic emission signal/feature extraction/recurrence plot/convolu-tional neural network/gated recurrent unit/deep learning分类
能源科技引用本文复制引用
刘忠,乔帅程,邹淑云,郑佳稳,吴怡恬..基于递归图和卷积神经网络-门控循环单元的水轮机空化状态识别方法[J].哈尔滨工程大学学报,2026,47(2):248-254,7.基金项目
国家自然科学基金项目(52079011) （52079011）
湖南省自然科学基金项目(2023JJ30032) （2023JJ30032）
湖南省研究生科研创新项目(CX20230903). （CX20230903）
