A Two-Stage Prediction Framework for Oil and Gas Well Production Based on Classification and Regression Models

Abstract

Oil and gas well production forecasting is a crucial aspect of exploration and development, yet it confronts challenges posed by complex geological conditions, incomplete data sets, and nonlinear interactions among multiple factors, all of which constrain the accuracy of traditional forecasting methods. Furthermore, existing approaches often overlook the intrinsic variations in production characteristics among wells due to differences in geological settings and development histories, employing generalized models that further hinder the enhancement of forecasting effectiveness. To address these issues, this study introduces an innovative staged forecasting framework that integrates classification and regression algorithms to achieve precise production forecasting for new oil and gas wells. Leveraging historical data, the framework utilizes a 3 year cumulative production as the label and establishes reasonable thresholds to categorize wells into low-yield and high-yield groups, thereby capturing the distinct production characteristics of each category. Subsequently, advanced classification algorithms are employed to train a classification model that accurately categorizes new wells. Dedicated regression models are then trained separately for the classified low-producing and high-producing wells, aiming to further elevate the accuracy of production forecasting. The application results demonstrate that the proposed method, compared to conventional forecasting approaches, exhibits significant improvements in both prediction accuracy and practicality, offering a novel perspective and methodology for the field of oil and gas well production forecasting.