Research And System Design Of Pumping Unit Working Condition Diagnosis Based On Deep Learning

The indicator diagram of the pumping unit can reflect the overall operation of the oil well,and accurately identifying of the indicator diagram is helpful to prevent and solve problems encountered in the oil production process.Traditional identification method of pumping crew is inefficient and has large error,while deep learning can directly learn the significant features of samples from the data set and identify the indicator diagram,which is a more effective method in fault diagnosis of pumping units.Therefore,this dissertation improves the existing algorithm and introduces the field of working condition identification of pumping units,so as to improve the accuracy of working condition identification of pumping units.Since the effect of traditional generation of countermeasure network enhancement samples is poor and the types of generated samples are less,this dissertation adds SKnet module and gradient reconciliation mechanism on the basis of it to improve attention to channels and achieve the purpose of enhance the indicator diagram data.In order to extract more effective features from the indicator diagram,this dissertation adds the SKnet module on the basis of auxiliary generation of countermeasure network.At the same time,in order to avoid over fitting,this dissertation also adds a gradient reconciliation mechanism to make the network more balanced when training based on indicator diagram data set.After multiple trials,the improved auxiliary generative adversarial network is better for data set augmentation.In order to improve the accuracy of fault diagnosis for pumping units,this dissertation studies the similarity of indicator diagrams based on deep learning and standardized Euclidean distance method,and obtains good experimental results based on this method.Then the similarity judgment is studied based on lightweight network and triplet loss.At the same time,on the basis of two experiments,proposes a new diagnosis method,which improves the accuracy of the condition diagnosis algorithm.Based on the above work,this dissertation designed a diagnosis system and tests and verified the whole system at the pumping site.The test results show that the system can not only collect data and accurately diagnose conditions,but also meet the real-time requirements.