A New Method of Using Production Data to Predict the Effect of Oil Well Liquid Lifting Based on Deep Learning

Abstract Liquid lifting is a low-cost, straightforward, and efficient measure for stabilizing production, achieved by adjusting liquid production to reduce bottom-hole flow pressure, thereby optimizing reservoir development and increasing recovery rates. However, traditional research methods for liquid lifting no longer meet the demands of contemporary oilfield big data applications. Particularly in the selection of target wells for liquid lifting, manual screening based on chart methods faces challenges such as low efficiency, high workload, and poor generalization capabilities. This study focuses on the oil wells in the high water cut phase of the SL oilfield in China and proposes a novel data-driven approach to predict the liquid lifting effects of oil wells. This method forecasts the effect of liquid lifting measures based on dynamic and static production data before the implementation of liquid lifting measures, facilitating intelligent selection of oil wells for liquid lifting measures. Firstly, each evaluation metric is assigned weights by utilizing the Coefficient of Variation-G1 Hybrid Cross-Weighting method. The comprehensive evaluation score is obtained through a weighted sum. Subsequently, the self-organizing map (SOM) clustering method is applied to categorize the oil well liquid lifting effects into four classes (A, B, C, D), representing different levels of effect. Additionally, a learning sample dataset is constructed by selecting multidimensional time-series features before liquid lifting. The oil well liquid lifting effect classification model is established using the Bi-LSTM sequence-to-label deep learning algorithm. Comprehensive testing validates the superior performance of the model, especially with classification accuracies of 0.96 and 0.83 on the training and testing sets, respectively. These results outperform other classification models such as LSTM, XGBoost, and SVM, providing an effective tool for intelligent oil well selection in liquid lifting measures.