Research On Feature Extraction And Recognition Of Pressure Pulsation Signal In Draft Tube Of Pump Turbine

Pumped storage power station,with its excellent peak shilling,frequency modulation and black start function,plays a more and more important role in the development of renewable energy power generation network in recent years.During the operation of pumped storage power station,the periodic vibration of the vortex rope pressure pulsation in the draft tube of pump turbine impinges on the wall of the draft tube,which may induce the excitation of the flow parts of pump turbine and lead to cracks in the turbine runner blades,posing a great threat to the operation and maintenance of the whole pumped storage power station.herefore,the feature extraction and intelligent recognition of pressure pulsation in the of the draft tube have always been the focus of research.This paper focuses on the pressure pulsation in the draft tube of a pump turbine,selects the measured data of the pressure pulsation in the draft tube o f a pumped storage power plant,and analyzes the pressure pulsation in the dr aft tube from three aspects:feature extraction,mode decomposition and intellig ent recognition and classification.Firstly,Ensemble Empirical Mode Decomposi tion(EEMD)and index energy method are used to extract the features of one-dimensional time domain signal of pressure pulsation,and the internal energy change is analyzed.Secondly,Using Empirical Mode Decomposition,Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)a nd Variational Mode Decomposition(VMD)methods were used to analyze the measured signal of pressure pulsation in the draft tube of a pumped storage p ower station,and then Hilbert Transform(HT)was used to uniformly represent the spectrum.The pressure pulsation of the draft tube with different amplitud e intensity is analyzed and compared.Finally,the Alexnet convolutional neural network was used to identify and predict the two-dimensional spectrum of pr essure pulsation in the draft tube of different intensities of EMD-HT,CEEMD AN-HT and VMD-HT,and the following conclusions were obtained:Firstly,the comprehensive analysis method of EEMD and index energy m ethod can effectively extract the characteristics of one-dimensional pressure pul sation signal in the time domain of the draft tube.The maximum and mean in dex energy can accurately reflect the energy change trend of the pressure puls ation signal.Secondly,through EMD-HT,CEEMDAN-HT and VMD-HT,it is found that the pressure pulsation signal of the draft tube with different amplitudes is mainly dominated by the low frequency vortex band pressure pulsation of about 3Hz,and there are some moderate and high frequency low amplitude pressure pulsation.In addition,compared with the other two decomposition methods,the VMD-HT algorithm can effectively solve the mode mixing problem and clearly show the main two-dimensional spectral characteristics of the pressure pulsation in the vortex band of the draft rope.Thirdly,the Alexnet convolutional neural network was used to identify and predict the two-dimensional spectrum of pressure pulsation of with different intensities under the above three decomposition algorithms.It was found that the CEEMDAN-HT decomposition method had the highest recognition accuracy of 83.33%for the weakest intensity pressure pulsation signal figures.For the moderate intensity pressure pulsation signal,the VMD-HT decomposition method has the highest recognition accuracy of 95.66%.For the highest intensity pressure pulsation signal image,the recognition accuracy of VMD-HT decomposition method is the highest,which is 87.27%.For the overall pressure pulsation signal figures,the VMD-HT decomposition method has the highest comprehensive recognition rate,and the recognition rate reaches 91.08%.Through comprehensive comparison,the VMD-HT analysis method has the highest recognition rate and the best recognition effect.This method has important engineering significance for the intelligent recognition and prediction of pressure pulsations in pumping storage power stations.