| With the rapid development of the Chinese economy and the continuous expansion of social demand for energy,hydropower,as a renewable energy source,is becoming increasingly important in the power system,with larger installed and unit capacities.The stable operation of the units is closely related to the safety of the power grid.However,cavitation is a common destructive phenomenon in hydraulic machinery,which can cause overcurrent components to fall off or erode,fluid to separate from the flow,reduce the efficiency of the unit,and affect the safe and stable operation of the unit.Therefore,effective detection and accurate identification of cavitation state of hydraulic turbines have important research value,which can avoid the longterm operation of the unit in severe cavitation state,extend the service life of the unit,and improve economic benefits.This article combines the modal decomposition algorithm in the field of signal processing with machine learning algorithms to study the modal feature extraction of cavitation signal in the time-frequency domain under different cavitation states,and establish a cavitation state recognition model.The main research work and results are as follows:(1)A signal acquisition software system and hardware acquisition equipment were developed to collect spindle vibration signals,tailpipe wall pressure pulsation signals,and acoustic emission signals at the guide vane elbow under different cavitation states.The rationality of the acquisition system was verified through simulation and hydraulic turbine experiments.Cavitation experiments were conducted on a hydraulic turbine in a power station in China.Sensors were installed at designated locations on the unit,and by adjusting the guide vane opening and speed,the efficiency of the hydraulic turbine was reduced to obtain the cavitation characteristic curve and collect four typical cavitation signals.(2)To evaluate the optimal feature extraction algorithm for cavitation signals,this article used conventional variational mode decomposition,optimized variational mode decomposition,and local mean decomposition algorithms to extract features of four typical cavitation signals.The optimized variational mode decomposition has better robustness,more thorough modal decomposition,and better decomposition accuracy,and can accurately extract intrinsic mode components.(3)With the example of spindle vibration acceleration,this article studied the frequency changes of cavitation signals,showing that as the cavitation coefficient decreases and the cavitation depth increases,except for the fundamental frequency,the low-frequency band frequency gradually increases.The highest frequency of the mode at the critical cavitation point is 6579Hz,the frequency of the high-frequency component in the initial cavitation process is 5214Hz,and the frequency of the high-frequency components at the cavitation aggravation point and severe cavitation are 6126Hz and 5866Hz,respectively,showing a trend of first increasing and then decreasing.When severe cavitation occurs in the hydraulic turbine,the 10th harmonic component of the fundamental frequency disappears,the number of signal modes decreases,and the frequency band span of each mode is large.(4)To classify the cavitation state of the hydraulic turbine,this article established conventional ensemble learning algorithm models,Bayesian optimization algorithm models,and CNN convolutional neural network models based on the theory of machine learning algorithms.The research results show that the accuracy of each algorithm is generally improved after optimization,with the accuracy of random ridge and AdaBoost algorithms exceeding 54%,and the accuracy of XGBoost algorithm increasing to 68.77%,an increase of 17.8%.The CNN convolutional neural network has a high recognition ability for cavitation aggravation and severe cavitation states,but a lower recognition ability for critical and initial cavitation. |
PDF Full Text Request