A Wax Deposition Prediction Method for Pumping Oil Wells Based on LSTM Neural Networks

Abstract

Wax deposition in oil wells reduces oil production efficiency, increases maintenance costs, and raises the occurrence rate of accidents. Traditional methods have difficulty accurately predicting wax deposition in oil wells, and oil fields typically adopt a passive response mode, resulting in decreased efficiency. Hence, this study proposed a wax deposition prediction method for oil wells based on the long short-term memory (LSTM) neural network to effectively extract and utilize the information content of historical dynamic monitoring data of oil wells and achieve real-time prediction and early warning for wax deposition. A predictive indicator system for predicting failures caused by wax deposition in pumping wells was established, and a wax deposition prediction dataset was prepared by integrating and normalizing collected operational data of oil wells and conducting sensitive parameter analysis. The Grey Wolf Optimization (GWO) algorithm was used to optimize the constructed LSTM prediction model, and the optimized LSTM model was trained and tested using data from the prediction dataset. The accuracy of the model was confirmed through experiments. Results demonstrate that the proposed LSTM-based wax deposition prediction method in this study achieves R-squared values of 0.8453 and 0.9439 on the test set before and after optimization, respectively, effectively improving the accuracy of wax deposition trend prediction in oil wells. This study can assist in predicting wax deposition trends in oil wells, enabling oil field workers to formulate preventive measures in advance, thereby reducing production costs and the occurrence rate of accidents.