Rapid Forecasting of Carbon Storage and Migration using Neural Operators and Transfer Learning

Speaker: Siddharth Misra, Ph. D., Associate Professor, Harold Vance Department of Petroleum Engineering, Texas A&M University
The combination of Fourier Neural Operator (FNO) network and Transfer Learning offers a powerful solution for reducing both computational time and data needed in predicting the 3D migration of CO2 plumes within a storage reservoir. This approach enables efficient and accurate analysis of CO2 saturation and pressure evolution in large reservoirs with diverse geological and engineering conditions. The effectiveness of this technique was demonstrated through experiments on the 3D SACROC geomodel, which showed that the transfer learning implementation requires 75% less training time and 50% less data compared to conventional approaches. A traditional forecasting of 40 minutes was reduced to 1 min using this approach.