| The sucker-rod pumping accounts for a large proportion in the oil extraction. Because of dispersed location and complicated circumstance of the oil wells and equipments, sometimes the diagnoses of sucker-rod pumping and fault treatment are delayed, which makes the diagnosis of sucker-rod pumping quite a difficult problem in the oil production field. Therefore, it is very important to diagnose the fault of the pumping units reliably and provide operation advices timely, which means a lot for improving the production efficiency and economical operation in oil field. The indicator card is the basis of analyzing the pumping condition which reflects the working condition of sucker-rod pumping unit intuitively and feature extraction is the key step for working condition diagnosis.1 This paper introduces the development and main method of fault diagnosis, describes the operation principle of pumping unit and the shaping process of indicator card. Then the characteristic and the normalization preprocessing for indicator card are discussed particularly. A set of indicator diagram samples is established.2 Three feature extraction algorithms of the indicator card are researched in the paper. Fourier Descriptors(FD) are adopted as the first algorithm; choosing characteristic moments of indicator card is another algorithm and the last one forms a gray matrix, calculating the gray statistics characteristics of the indicator card. Then they are compared in terms of time and complexity.3 As a sort of statistical learning method, Support Vector Machine (SVM) is applied to indicator card classification. Basic theory and methods of SVM are described. The samples are trained by Support Vector Classification and researches are done on the influences of the selection of kernel functions and parameters of SVM. The three feature extractions algorithms are compared according to the simulation results. Experimental results show that the correct recognition rate is above 95%. |
PDF Full Text Request