Innovative Machine Learning Approach to Locate the Remaining Gas

Abstract Locating remaining gas in a producing gas field poses several challenges, primarily due to complexities in reservoir dynamics, data availability, and technological limitations. Gas reservoirs often exhibit complex geological formations and heterogeneities. Variations in porosity, permeability, and rock types can create irregular gas distribution, making it difficult to accurately predict where remaining gas pockets might be located. Moreover, sparse well coverage, incomplete data sets, or inaccuracies in historical data can hinder accurate reservoir modeling and forecasting. Addressing these challenges requires a multidisciplinary approach combining geological understanding, advanced reservoir engineering, and innovative technologies. We designed a new approach that predicts the reservoir properties via Machine Learning algorithm, Neural Network Inversion, in order to locate the remaining reserves. In this workflow, we predict 3D volumes of porosity and water saturation using deep learning neural network algorithm. Then, we resampled these volumes into reservoir 3D structural grid. Then, we defined the net and calculated the gas in-place for different reservoir segments. And finally compare these calculations with the dynamic production data. The proposed workflow was calibrated through multiple blind well tests and the predictions showed high accuracy. The results of the project were astonishing. 500+ billion cubic feet of gas were added to the portfolio in the form of infill and near-field exploration opportunities. Consequently, a next phase of development was proposed based on the findings. The proposed approach demonstrates that the data integration and advanced predictive modeling are key strategies to optimize gas recovery and extend the productive life of gas fields.