Research On Fault Diagnosis Method Of Pumping Unit Based On Domain Adaptation

Oil is a very important resource,which is related to national security and social stability.The pumping unit is a kind of mechanical equipment in the petroleum exploitation project.It is widely used in the oil field due to its simple structure and low cost.With the continuous development of oilfield production,most of the old oilfields in my country have entered a period of high water cut,the oil production environment of the pumping unit has become more and more complex,and failures occur frequently,affecting the production efficiency and production safety of the oilfield.Therefore,it is of great engineering significance to establish an accurate and effective pumping unit fault diagnosis model.The indicator diagram data of pumping units collected under different conditions have different distribution,which violates the requirement of the same distribution of the conventional machine learning model data.Therefore,this paper studies the cross domain fault diagnosis method of pumping units based on domain adaptation in migration learning,combined with the industrial characteristics of pumping units.It mainly includes the following contents:1.Aiming at the noise problem of pseudo-labels in traditional fine-grained domain adaptation methods,a fault diagnosis method for pumping units based on structural information preserving domain adaptation network is proposed.The method divides the source and target domains into multiple subclasses according to the number of fault types,and then achieves unsupervised domain adaptation by adjusting the distribution of relevant subclasses in the cross-domain feature layer.This method adopts the fuzzy clustering method in the generation of pseudo-labels in the target domain,and the obtained pseudo-labels can avoid some label noise while retaining the structural features of the target domain data,which can effectively improve the efficiency of domain adaptation.Experiments are conducted on real production datasets from different sources,and the proposed method achieves an average accuracy of94.69%,outperforming conventional methods.2.Aiming at the fact that most traditional fine-grained domain adaptation methods only focus on category information,and the conventional methods are not related to the difference optimization between injection subclasses,a fault diagnosis method of pumping unit based on deep substructure domain adaptation network is proposed.This method discards the category labels of the data itself,uses the clustering method to generate more substructure labels,divides the samples of the two domains into multiple substructures,and jointly optimizes the differences within and between the substructures,in which the differences between the substructures are maximized,so as to obtain the features with more obvious classification boundary.In the cross domain experiment,the average accuracy of the proposed method is 99.67%,which further improves the performance of the model.3.Aiming at the problem that the fault types of pumping units are different in different scenarios,resulting in the negative migration of conventional domain adaptation methods,a pumping unit fault diagnosis method based on adversarial partial domain adaptation is proposed.This method calculates and adds the transferability weight of the sample in the process of adversarial domain adaptation,so that the weight of the source domain common category sample is large,while the private category sample has a small weight,so it can reduce the source domain private category sample to the domain adaptation process.Impact.In addition to this,weighted contrastive domain differences are added to facilitate separation between different categories and obtain better classification features.In some domain adaptation experiments,the proposed method achieves an average accuracy of 79.13%,which can effectively suppress negative transfer and improve the diagnostic ability of the model.