Research On Intelligent Diagnosis Method For Well Condition Of Traveling Beam Pumping Unit Based On Measured Electric Parameters

At present,intelligent algorithm has been widely used in the field of fault diagnosis of pumping wells.Most of the existing diagnosis models are based on the measured indicator diagram.Because of the poor measurement accuracy and high cost of indicator diagram,most of the diagnosis models based on indicator diagram are unsatisfactory in practical application.Because the fault features extracted from electrical parameters are not separable,the diagnosis effect based on electrical parameters is still not effective.Therefore,it is of great engineering significance and economic value to study the fault diagnosis technology of pumping wells based on electrical parameters.In view of the existing problems and deficiencies in the current research,this paper mainly includes the following aspects:Through the research and analysis of each part of pumping well system,the simulation models of the pump indicator diagram,indicator diagram and electric power curve under various working conditions are established.The input power curve of the motor and the corresponding hanging-point indicator diagram are obtained by using the model,and the training sample database of the system dynamic simulation is established;Establish the simulation models of motion analysis and dynamics of ground machinery.According to the relationship between energy parameters of the ground nodes,the influence factors and sensitivity of the indicator diagram inversion are analyzed.Based on the corresponding relationship between electrical parameters and indicator diagram,the sample database is established from two technical routes,and the intelligent inversion model of indicator diagram and the analytic model of indicator diagram are established based on the improved threshold fitting method.The two models are analyzed and compared with the test data,and the accuracy of the model is verified with the measured data;Establish the model of sample numerical feature extraction,based on the characteristics of sample data and application conditions,the numerical characteristics of samples are extracted reasonably.Draw the electric power curve and the suspension pointdisplacement diagram,and generate the binary image from the diagram and the hanging-point indicator diagram,and then extract the graphic features of the binary image by establishing the graphic feature extraction model.The sample curve is decomposed by wavelet packet,the time-domain signal is transformed to frequency-domain,and then the time-frequency characteristic parameters are extracted.Finally,the extracted feature parameters are extracted by principal component,and the final feature samples are obtained;The similarity measure classifier,BP neural network classifier and RBF neural network classifier are established respectively,and then the structure parameters of neural network classifier are optimized.According to the diagnosis characteristics and results of each classifier,the condition recognition model of parallel recognition of multiple classifiers is established.And the accuracy of each classifier and the general condition recognition model is verified by the measured data.