Data Augmentation Method Based on Improved Generative Adversarial Network for the Sucker Rod Pump System

摘要

In this paper, a novel data augmentation method is proposed for imbalanced fault diagnosis in the sucker rod pumping system (SRPS) based on the improved generative adversarial network (GAN). The dynamometer cards (DCs) of minority fault classes are expanded steadily through learning the data distribution of the original imbalanced training data. The generalization ability and accuracy of the different fault diagnosis models are improved with the expanded DCs. Firstly, benefit from introducing the auxiliary conditional information to the Wasserstein GAN with gradient penalty (WGAN-GP), a stable and practical framework is constructed to generate specific categories of DCs. Secondly, the traditional random noise is replaced by the Gaussian mixture noise to guarantee the diversity of the generated DCs with few training samples. Moreover, problematic training, such as unstable training and mode collapse, is effectively avoided with the presented optimization strategy, training structure and data processing method. Eventually, generated data evaluation and comparative experiments are conducted using practical DCs collected from several oil wells with similar working conditions in the northeastern of China. The results verify the validity of the proposed data augmentation method.