Application of an artificial neural network in predicting the effectiveness of trapping mechanisms on CO2 sequestration in saline aquifers

Predicting the effectiveness of geological CO2 storage and evaluating the field application of successful CO2 sequestration require a large number of case studies. These case studies that incorporate geologic, petrophysical, and reservoir characteristics can be achieved with an artificial neural network. We created an artificial neural network model for geological CO2 sequestration in saline aquifers (ANN-GCS). To train and test the ANN-GCS model, data of residual and solubility trapping indices were generated from a synthetic aquifer. Training and testing were conducted using Python with Keras, where the best iteration and regression were considered based on the calculated coefficient of determination (R-2) and root mean square error (RMSE) values. The architecture of the model consists of eight hidden layers with each layer of 64 nodes showing an R-2 of 0.9847 and an RMSE of 0.0082. For practical application, model validation was performed using a field model of saline aquifers located in Pohang Basin, Korea. The model predicted the values, resulting in an R-2 of 0.9933 and an RMSE of 0.0197 for RTI and an R-2 of 0.9442 and an RMSE of 0.0113 for STI. The model was applied successfully to solve a large number of case studies, predict trapping mechanisms, and optimize relationships between physical parameters of formation characteristics and storage efficiency. We propose that the ANN-GCS model is a useful tool to predict the storage effectiveness and to evaluate the successful CO2 sequestration. Our model may be a solution to works, where conventional simulations may not provide successful solutions.