Working Condition Diagnosis of Sucker Rod Pump Based on Kernel Density Distance Domain Adaptation

Due to significant distribution differences in motor power curves(MPCs)signals across various models of sucker rod pump system (SRPS) motors, a shortage of labeled MPCs samples for diagnosis, and poor generalization of diagnostic models, this paper proposes a method for diagnosing the working conditions of SRPS based on a multi-kernel density domain adaptive network model (MKDDAN). First, a one-dimensional convolutional neural network is employed to extract the features of labeled MPCs from the source domain and unlabeled MPCs from the target domain. Then, a multi-kernel density estimation (MK-DE) strategy is utilized to reduce the conditional distribution differences between the source and target domain data features and to provide pseudolabels for the target domain data. Subsequently, multi-kernel maximum mean discrepancy is used to align the marginal distributions of the two domains, resulting in a diagnostic model for SRPS working conditions. Finally, the method is tested on a dataset of MPCs measured from an actual oil field through six transfer tasks. The results demonstrate the feasibility of MKDDAN in cross-domain fault diagnosis for SRPS, indicating that transfer learning theory has promising research potential in the diagnosis of SRPS working conditions.