Machine Learning Prediction of Gas Hydrates Phase Equilibrium in Porous Medium
Saeed Beheshtian, Sara Kishan Roodbari, Hamzeh Ghorbani, Mohamadreza Azodinia, Mohamed Mudabbir
Abstract
Gas hydrate (GH) reservoirs play an important role in the economy of countries and various industries depend on it. Since these reservoirs display specific behavior, understanding their thermodynamic aspects can help engineers to find their behavior. In this study, 1005 data points were collected from an Iranian gas reservoir for the construction and development of three novel machine learning techniques: Extra Tree (ET), Gene Expression Programming (GEP), and Light Gradient Boosting Machine (LightGBM) to predict gas hydrate pressure (P<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">H</inf>). From the total data, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$70{\rm{\% }}$</tex> was utilized for constructing artificial intelligence algorithms, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$15{\rm{\% }}$</tex> for testing, and another <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$15{\rm{\% }}$</tex> for validating outputs. Based on analyzing the algorithm results and comparing statistical errors, it is concluded that the LightGBM algorithm exhibits higher performance accuracy compared to other algorithms (RMSE = 1.875 and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R2 = 0.97$</tex> ). The LightGBM algorithm has several advantages, including high accuracy, high speed, and high efficiency, as well as a histogram-based approach that enhances performance accuracy through short memory retention. This algorithm utilizes parallel computations, making it suitable for large datasets. Through gradient-based learning, LightGBM captures complex relationships and provides accurate predictions, even with nonlinear patterns. Additionally, it seamlessly handles classification features, eliminates preprocessing needs, and limits excessive internal regularization. Its ability to determine matching metaparameters and handle unbalanced data enhances its performance.