Prediction and classification of faults in electric submersible pumps

Abstract

As a core component of oil production equipment, electric submersible pumps (ESPs) have been widely used in offshore oil fields to improve oil well production. There is an urgent need for an effective method of predicting and classifying ESP faults in advance because the traditional approach of diagnosis after a fault occurs results in serious economic losses. This paper describes a method for the prediction and classification of ESP faults, combining a backpropagation neural network with artificial feature extraction. To overcome the influence of noise and the small sample size of fault data, overlapping sampling and manual feature extraction are applied to the original data. To predict the occurrence of faults in advance, the fault events are divided into three stages. Experimental results show that the accuracy of fault prediction using samples with feature extraction applied is better than that using only the original samples. Compared with other algorithms, the proposed method obtains better prediction and classification results, demonstrating that artificial feature extraction is indispensable and effective in the process of fault prediction.