An Automatic Data Augmentation Method for Working Condition Diagnosis of Rod Pumping Systems Based on Teacher Knowledge

Abstract

Working condition diagnosis is an important means of evaluating the operating state of rod pumping systems. As the data source of working condition diagnosis, the quality of indicator diagrams will have a significant impact on the diagnosis results. In the actual oil field production process, the number of samples between indicator types is usually unbalanced, so it is an important means to improve the diagnostic accuracy by using data augmentation methods. However, traditional data augmentation methods require manual design, and the experimental results are not satisfactory. We propose an automatic data augmentation method based on teacher knowledge for working condition diagnosis of rod pumping systems. This method adopts an adversarial strategy for data augmentation and optimization and uses the teacher model as prior knowledge to generate information-rich transformation images for the model, thereby improving the generalization of the working condition diagnosis model. Specifically, our method makes the augmented images adversarial to the target model and recognizable to the teacher model. Compared with traditional methods, this method can automatically select the correct data enhancement method according to different indicator diagram sample sets to solve the corresponding problems. Our method has an accuracy of more than 98% in the diagnosis of actual oil field operating conditions. The experiment showed that the accuracy of this method was more than 5% higher than that of the traditional data augmentation methods in the task of condition diagnosis, which shows that this method has research and development value.