Research On The Feature Extraction And Recognition Method Of Flow States In The Vaneless Space Under Steady State Condition Of Prototype Pump Turbine

In recent years,with the expansion of renewable energy power generation,the demand for energy storage technology for power grids has continued to increase.The pumping storage power station has a variety of functions such as peak-regulation,frequency adjustment,black start,and emergency accidents,which is the largest and most mature energy storage method today.Pump turbine is the core component of energy conversion in pumped storage power station.It is very important to ensure its stable operation for the safety of pumped storage power station.Affected by the influence of Rotor-stator interaction,unsteady flow state is easy to occur in the vaneless space of pump turbine,which produces abnormal pressure pulsation and threatens the safety and stability of pump turbine operation.Therefore,in order to ensure the stable operation of pump turbine,it is very necessary to study the method of features extraction and recognition of flow states in the vaneless space so as to monitor the flow state.In this paper,the pressure fluctuation signal in the vaneless space of prototype pump turbine is taken as the research object,and the feature extraction and recognition method of the vaneless space flow states are studied from three aspects:mode decomposition,feature calculation and intelligent recognition.First,an improved empirical wavelet transform using the least squares method and mathematical morphology is proposed.Then,based on the improved empirical wavelet transform,energy feature vector and convolutional neural network,a set of innovative feature extraction and recognition method for flow states in the vaneless space is proposed.Finally,the proposed method is tested by using the measured pressure fluctuation signal in the vaneless space of prototype pump turbine.The main research conclusions of this paper are as follows:Firstly,the improved empirical wavelet transform method is used to decompose the pressure fluctuation signal in the vaneless space of pump turbine,which avoids the mode aliasing phenomenon in the empirical mode decomposition method,while removing the trend term in the signal,improving the accuracy of time-frequency analysis,and makes up the shortcoming that the traditional empirical wavelet transform cannot correctly identify the main frequency in the signal due to the influence of environmental factors,and effectively improves the decomposition effect.Secondly,the feature vectors obtained based on the improved empirical wavelet transform and energy feature vector method can accurately reflect the characteristic frequency distribution in the pressure fluctuation signal,accurately extract the characteristics of different flow modes,and serve as the input of the convolutional neural network.Thirdly,the intelligent flow states recognition model obtained by the method proposed in this paper effectively realizes the feature extraction and recognition of pressure fluctuation signals in the vaneless space of pump turbine under the conditions of power generation,pumping and idling.The average accuracy was 99.15%.