Summary: We developed a learning-based inversion-free prediction method to enable real-time forecast of geological carbon storage.

Accomplishment: The traditional inversion-based iterative model-data integration greatly limits geologic carbon storage forecasts. A new learning-based inversion-free framework (LIP) developed by ORNL researchers avoids computationally expensive and challenging model inversion, enabling to efficiently solve high-dimensional complex model predictions. This is accomplished by developing a robust machine learning model to effectively learn the relationship between observation and prediction variables based on their forward model simulation samples and then inferring prediction and its uncertainty from the relationship. The developed LIP framework has potential to fundamentally change how real-time decisions are made about geologic carbon storage operations. LIP is computationally efficient which only requires a few hundreds of fully parallelizable forward simulations. Additionally, LIP provides continually updating forecasts for CO2 distributions from streaming observations, thus providing operators with earlier warning of off-normal behavior and more time to implement mitigation measures.

Reference: https://doi.org/10.3389/fenrg.2021.752185