A Study On Condition Monitoring And Diagnosis Of Sucker-rod Pumping System Based On SVM

Sucker-rod pumping system dominates the main position in the petroleum production of the current world, especially during the tail development of oilfields, to understand and master working-pump’s condition of oilfields plays an important role in the exploit oil well. This paper introduces a fault diagnosis method based.on support vector machine (SVM), using its good recognition ability to small sample problems to diagnose the fault under the oilfields.(1) This paper thinks the sucker rod pumping wells as the research objects to deeply analyze the indicator diagram and it’s formation principle.(2) Then aiming at indicator diagram of pretreatment and normalization, using two methods to extract the indicator diagram’s eigenvector of image moment invariants and wavelet packet-energy entropy (WP-EE).(3) Discussing the important theory of statistical learning theory and SVM, they are VC dimension, promotional boundary and structural risk minimization principle. Then introduces some algorithms of linear, nonlinear and classification based on support vector machine.(4) Based on the above, researching the method to identify the indicator diagram classification based on SVM. Through comparing the final results by different normalization methods, different kernel functions and different parameters’ combination, we selected the [0,1] as normalized way, RBF kernel function and the best parameters<c, g> combination are selected in the finally, the classification accuracy can achieve98%.Theoretical analysis and experiments show that support vector machine in the sucker-rod pumping system’s fault diagnosis has a better ability to identify with a certain practicality.