Study on Working Condition Prediction of Suck Rod Pumping System Based on Long Short-Term Memory Network

Due to the complexity of downhole conditions, the fault diagnosis and prediction of suck rod pumping system have always been a hot issue of concern for oilfield staff. Currently, many researchers have done a lot of work on theoretical methods and practical applications of fault diagnosis, but there are few studies on fault prediction. Aiming at this problem, this article adopts the invariant moment theory to extract the eigenvalues of the dynamometer card, uses the eigenvalues of the dynamometer card as the prediction variables of the model. And achieves the forecasting of the changing trend of the working state of the suck rod pumping system through establishing a long short-term memory network prediction model. The results of the example analysis show that Mean Absolute Percentage Error is 1.144%, and the Mean Absolute Percentage Error is within the allowable range. This article presents a preliminary study on fault prediction for suck rod pumping systems, which has some guidance for practical application.