Study On Condition Diagnosis Of Electric Parameters Of Pumping Wells Based On Machine Learning

At present,there are more than 400000 production wells in China,of which more than 80% use pumping units as artificial lift.Accurate diagnosis of production conditions of pumping wells is the core technology of stabilizing single well production,prolonging pump inspection period,saving energy and reducing consumption.At present,there are many problems in condition diagnosis based on indicator diagram,such as high investment cost,data drift and distortion.In the production process of pumping wells,the electric parameters have the advantages of low acquisition cost,stable data and long-term continuous measurement.At the same time,there are many defects such as irregular data and many influencing factors.It is difficult to achieve accurate condition diagnosis by mathematical modeling alone.In this paper,firstly,based on the consideration of motor power loss,belt power loss,gearbox power loss and other factors,the mechanical transmission model method is applied to establish the mathematical and physical model of electric power to solve the net torque of crank shaft.Through this model,it is clear that there is a change conversion coefficient between electric power and net torque of crank shaft.Secondly,in the actual calculation of the conversion coefficient,due to the complex load change of the motor,it is difficult to solve the problem by using the mathematical and physical model.In this paper,the method of machine learning is used to establish a depth neural network model to solve the relationship coefficient of the electric power inversion of the crank shaft net torque,through which the conversion coefficient can be accurately solved.Then,with the help of the inversion of electric parameters,considering the structural characteristics and motion principle of the pumping unit,combined with the structural parameters and balance parameters of the pumping unit,the theoretical model of calculating the hanging point indicator diagram of the crank shaft net torque is established by using the mathematical and physical methods,and the prediction of the indicator diagram is realized by the model.Finally,using the calculated indicator diagram and convolution neural network algorithm,the structure of lenet5 convolution neural network is improved,and the working condition diagnosis model of indicator diagram of pumping well is established to complete the working condition diagnosis of pumping well.In this paper,through the study of mathematical and physical model,combined with big data machine learning method,a set of condition diagnosis technology based on electrical parameters is formed.Using this technology,six wells in D oilfield are calculated,and the corresponding indicator diagram of each well is obtained.The average error of the maximum load is2.38%,and the average error of the minimum load is 7.65%.The average coincidence rate of indicator area is 91.52%.The coincidence rate of condition diagnosis was 83.3%.Under the current situation of low oil price,the technology has broad application space.