oa Prediction Method for Carbon Sequestration Capacity in Depleted Gas Reservoirs Based on Artificial Intelligence Algorithms

The study focuses on developing a prediction method for carbon sequestration capacity based on artificial intelligence algorithms tailored to different types of depleted gas reservoirs. By simply inputting the highly relevant geological, fluid, and engineering parameters, the method can predict the carbon sequestration capacity for various types of gas reservoirs, providing a basis for selecting suitable geological sites for carbon sequestration. Initially, by investigating existing instances of carbon sequestration in depleted gas reservoirs, a series of data pertaining to geology (such as reservoir porosity, permeability, effective thickness, interlayer heterogeneity, etc.), fluids (diffusion coefficient, water-rock reaction coefficient, water saturation, etc.), and engineering parameters (injection pressure, injection rate, etc.) are acquired. Subsequently, the mechanisms of carbon sequestration in different types of gas reservoirs are clarified, and datasets with high relevance to the carbon sequestration mechanisms are selected for feature selection and correlation analysis to prevent model overfitting. Thirdly, different artificial intelligence algorithms (including neural network models, XG Boost, etc.) are trained using the training set for different depleted gas reservoirs, and the optimal artificial intelligence algorithm is selected through the validation set. Finally, more samples are used for training in the later stages to ensure the stability of the data-driven model.