Research On Intelligent Diagnosis Method And Software Development Of Pumping Well Working Condition

Intelligent diagnosis of pumping well working conditions plays a vital role in the process of oilfield exploitation.At present,the diagnosis method of pumping well working condition only uses the indicator diagram method,ignoring other characteristics of the oil well,resulting in the limited classification ability of the diagnosis model and the low accuracy of the model.For this reason,this paper has carried out the following research on the method of intelligent diagnosis of pumping well conditions.Taking into account the changes in the acting point of the frictional force caused by the axial movement of the sucker rod,a simulation model was established based on the static and dynamic coordinates for the longitudinal vibration of the sucker rod string,the suspension point indicator diagram,and the pump indicator diagram under different working conditions.A dataset of pump power diagrams under various operating conditions has been established through extensive simulation calculations,providing training samples for the diagnostic model of pumping well operating conditions.The longitudinal vibration simulation model of sucker rod string in directional wells is established based on the continuous system wave equation.With the displacement and load time series of the measured dynamometer card as the boundary conditions of the definite solution,the numerical simulation models are established based on the finite difference method and the numerical integration method,respectively,to realize the transformation from the measured dynamometer card in directional wells to the pump dynamometer card.This laid the foundation for establishing a diagnostic model based on pump indicator diagram.A diagnostic model for pumping well operating conditions based on convolutional neural networks was established,and the model parameters were optimized through training on the downhole pump power graph dataset.The training results show that the graphical features extracted based on Res Net18 convolutional neural network and the established diagnostic model have high diagnostic accuracy.A training sample for the diagnosis model of pumping well operating conditions was established by comprehensively applying the graphical features of pump power diagram and ground power diagram feature parameters.Based on the light GBM ensemble learning algorithm,a Res Net-light GBM pumping well condition diagnosis model was established.Compared with the Res Net18 convolutional neural Network tomography diagnosis model,the dimension of characteristic parameters of the Res Net-light GBM pumping well condition diagnosis model has increased,and the model expression ability has improved,further improving the model classification accuracy.Developed computer software for "Diagnosis of Pumping Well Working Conditions".The software has been verified to have good diagnostic accuracy through pump power diagram test samples,which can meet the requirements of practical engineering applications.