FEATURE EXTRACTION AND IDENTIFICATION OF FLOW STATUSES IN VANELESS AREA OF PUMP TURBINE BASED ON EMPIRICAL WAVELET TRANSFORM AND CONVOLUTIONAL NEURAL NETWORK

It is very important to monitor and identify the flow status in the vaneless area (VA) of prototype pump turbine to ensure the operational safety and stability of pumped storage power station. In the present paper, a feature extraction and identification method of flow status is proposed based on the principles of empirical wavelet transform (EWT), dispersion entropy (DE) and convolutional neural network (CNN). Firstly, the EWT is used to decompose the pressure fluctuation signal. Then, the features of flow status are extracted by calculating the DE value of each component. Finally, by using feature-label pairs to train the CNN, the intelligent identification model is obtained to realize the identification of flow status in the VA. The real pressure fluctuation signals collected from a prototype pump turbine in the generating, pumping and idling modes are used to verify the proposed method. The average accuracy of the test is 94.84%, which proves the effectiveness of the method.