Engineering Applications of Artificial Intelligence to Forecast Production of Shale Wells

Abstract This study investigates the use of artificial intelligence (AI) and supervised machine learning techniques for predicting production from unconventional shale wells, utilizing actual field data collected over two years. While traditional methods like decline curve analysis (DCA) provide useful insights, they often overlook the complex factors influencing productivity and rely heavily on empirical equations. In this research, an AI-based Shale Analytics approach is employed, using Big Data Analytics to uncover unique patterns in field measurements. This enhances the evaluation and quantification of productivity factors and allows for better comparison. The analysis covers over 400 shale wells in the Marcellus Shale, emphasizing the influence of field measurements on well performance. TensorFlow and Keras machine learning libraries were used to create an Artificial Neural Network (ANN) tailored for extensive field measurement datasets, including variables like well metrics, hydraulic fracturing information, formation properties, completion designs, and operational data. The ANN learns from this vast dataset independently, without depending on prior models. The development process included organizing the datasets for training, calibration, and validation, enabling accurate predictions of production efficiency. The findings reveal that production performance can be effectively predicted using a detailed dataset of actual field measurements. The ANN model identifies optimal parameters for high predictive accuracy, indicating that AI integration in reservoir management could enhance both operational efficiency and accuracy, thereby improving industry practices.