Unsupervised learning-driven insights into shale gas Reservoirs: Production prediction and strategic applications

Abstract

Accurate and effective production prediction of oil and gas wells is of great significance for formulating reasonable and effective development strategies in the future. However, traditional empirical, physical and machine learning methods often require the use of labeled samples to predict production, which limits the performance of the model. Meanwhile, the oil and gas extraction industry remains in a state of sustained prosperity, with a significant number of as-yet-undrilled oil and gas wells (i.e., unlabeled samples) present in numerous fields. Therefore, this paper proposes an innovative framework based on unsupervised learning algorithms, called Unsupervised Production Prediction Framework (UPPF), aiming to use unlabeled well data for production prediction. In this study, the framework is applied to production example wells in the Sichuan Basin, using geological and engineering parameters of 240 wells for production prediction. A comparison of the prediction results between the UPPF framework and classic unsupervised learning methods demonstrates that the proposed UPPF framework can capture potential production patterns and features from unlabeled data, and performs well in predicting cumulative production of oil and gas wells. This innovative framework provides an advanced and feasible method for production prediction in oil and gas wells, providing strong support for decision-making and optimization in the field of oil and gas engineering. The results of this study are of great significance for promoting the development of production prediction methods and can be applied in similar fields.