Study On Cavitation State Recognition Method Of Hydraulic Turbine Based On Chaos Theory And Deep Learning

Hydroturbine is the main equipment of hydropower station.Cavitation is one of the common problems in the operation of hydroturbine at home and abroad,and it is also a basic scientific problem to be solved urgently in hydroelectric power generation.When cavitation occurs in the hydroturbine,high frequency pulse and radiated noise will cause water pressure fluctuation,induce the decline of energy characteristics of the hydroturbine,and lead to the deterioration of stability.When it is serious,it will cause the surface damage of the flow passage and shorten the life of the components.Therefore,it is of great significance to monitor and identify the cavitation state of hydroturbine.In this thesis,chaotic theory of the nonlinear dynamics and deep learning theory in the field of machine learning are introduced,and the cavitation state of hydroturbine is monitored based on the detection technology of Acoustic Emission(AE).The main research work is as follows:(1)A series of cavitation tests on a model Francis turbine was conducted,and the AE signals of typical measuring points under different cavitation conditions were collected.In order to improve the signal-to-noise ratio to analyze the dynamic characteristics of AE signals accurately,Empirical Mode Decomposition(EMD)threshold denoising algorithm was used to process the experimental AE signals,and the denoised AE signals were obtained.(2)In view of the strong nonlinearity and nonstationarity of acoustic emission signal from hydraulic turbine under cavitation,a chaotic analysis method of acoustic emission signals was established,and the chaotic characteristics and the change law of acoustic emission signals from Francis turbine under cavitation were studied.The results show that DC and LLE can describe the cavitation state qualitatively and the cavitation degree quantitatively.(3)Aiming at the problem that the recognition method of turbine cavitation state relies on manual feature extraction and has low robustness,a model of MCNN-DLSTM based on Convolution Neural Network(CNN)and Long-Short Term Memory(LSTM)network was proposed.The results show that compared with Bidirectional Long-Short Term Memory(BiLSTM)network and LSTM-CNN,the proposed model has higher recognition accuracy and robustness.The proposed mode is capable of real-time monitoring of hydroturbine cavitation state.In this thesis,a new idea for the study of cavitation monitoring and recognition of hydroturbines is provided.The relevant research results of this study provide a strong theoretical basis and technical support for the realization of the monitoring and recognition of the cavitation state of hydroturbines.