Automated history matching for CO2-ECBM reservoirs using Fourier-UNet surrogate models and Ensemble Smoother with Multiple Data Assimilation

In the development and application of CO2-Enhanced Coalbed Methane (CO2-ECBM) technology, reservoir simulation is critical for evaluating risks. However, accurately representing the complex physical processes in CO2-ECBM systems remains challenging due to dynamic behaviors within coal seams, and the high computational cost of high-fidelity simulations. Simplified geometric structures and the omission of micro- scale features further hinder precise modeling, while reliance on empirical parameters introduces uncertainties that may lead to discrepancies between predictions and observations. To address these challenges, this study proposes an automated history matching workflow that integrates a novel forward model and practical sensitivity and uncertainty analysis. The forward model, which combines AutoencoderKL and Fourier-UNet, achieves a balance between accuracy and computational efficiency, improving reconstruction accuracy by 59% and prediction accuracy by 61% compared to baseline models. Additionally, data assimilation is implemented using the Ensemble Smoother with Multiple Data Assimilation (ES-MDA) algorithm, which effectively refines model parameters and reduces parameter variance by 15.84% post-assimilation. The proposed workflow serves as a robust framework for optimizing CO2-ECBM project designs and improving risk assessment, offering significant potential for practical engineering applications.