An evolutional SVM method based on incremental algorithm and simulated indicator diagrams for fault diagnosis in sucker rod pumping systems

Abstract

Abstract Fault diagnosis of the sucker rod pumping system (SRPS) is a challenging problem in the oil industry. Recently, the computer-aided indicator diagram (ID) recognition techniques are becoming useful measurements to help engineers monitor the wells. However, the slow variation of SRPS state makes it difficult for a well to experience all the fault conditions. Using the IDs collected from other oil wells as training data will lead to a large difference between the distribution of training data and target data because of the interference information including noise and system operation characteristics in IDs. Moreover, the different occurrence probability of each fault leads to the imbalance of training data, therefore the fault tolerance of the diagnosis models will result in lower accuracy in identifying the fault IDs with less training samples. In addition, due to one-off training, the parameters of diagnostic model are always fixed resulting in lack of adaptive ability. To address this issue, an evolutional support vector machine (SVM) method for SRPS diagnosis is proposed. First, to obtain balanced training data with similar distribution to the target data, a novel model describing the working process of SRPS under various fault conditions is established. Subsequently, the static apparent stiffness feature and its extraction algorithm is proposed to fully retain the fault information of SRPS in IDs. Then, the incremental SVM is used to construct the diagnosis model based on generated IDs. At last, the proposed method is verified experimentally through the system parameters and IDs of many wells collected from oilfields, and then some conventional techniques are employed in the comparison studies. The obtained results show that the accuracy of the diagnosis model trained by simulated IDs is 11.2% higher than that trained by collected IDs. Besides, the parameters of proposed diagnosis model can be continuously evolved to improve the diagnosis accuracy and generalization ability. Furthermore, the proposed diagnosis model also has advantages in training efficiency and memory consumption.