Browsing by Author "Tavakolian, Mohsen"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Modeling of Methane and Carbon Dioxide Sorption Capacity in Tight Reservoirs Using Machine Learning (ML) Techniques
    (2023-01-19) Tavakolian, Mohsen; Kantzas, Apostolos; Hejazi, Hossein; Shor, Roman
    This work examines different machine learning methods, from shallow to deep learning. It investigates their capability to model 489 sets of experiments with 3806 data points where methane (CH4) and/or carbon dioxide (CO2) sorption capacity of shale and coal have been measured at different reservoir conditions. High-accuracy “integrated” shallow and deep learners/predictors were developed, without splitting the dataset into several batches. The best predictors were Random Forest Regressor (RFR) and Artificial Neural Network (ANN) with more than 98% and 96% accuracy, respectively. The developed models worked well with different combinations of potential features, e.g., gas (CH4 and CO2), rock type (either shale or coal family), TOC (total organic carbon), moisture, temperature, porosity, geological properties, etc., resulting in high accuracy (96 to 98%) for 69 sets of “unseen” experimental data with 1384 data points. Furthermore, a sensitivity analysis of the developed predictors showed their alignment with technical expectations of geology and reservoir engineering aspects. Of the best learners, RFR is simpler to apply and quicker to run with higher accuracy and ANN predicts 2% less accuracy on average. However, it operates with less risk of memorizing the data and a higher capability of accurately predicting unseen data. Both models are capable of being applied to lab scale and further developed for their application in reservoir scale simulators. In the end, the developed models are benchmarked against commonly used empirical correlations and pros and cons of each approach are discussed. A new correlation was also developed for CO2 phase change stage and added to this benchmark study.