This approach demonstrates improved CO2 plume predictions in geological sequestration, indicating enhanced monitoring capabilities.
Key Points
The hybrid cnn-transformer model achieved high accuracy with an average dice score of 0.799 ± 0.181 and substantial speedup in predictions.
Utilizing 3D CNNs for geological property processing and transformer blocks for dynamic sequences, the model adeptly captures CO2 plume evolution patterns.
With nearly 10,000 times acceleration on GPU, this model provides significant computational advantages over traditional reservoir simulations.
The method highlights the necessity for rapid detection tools in seismic monitoring, promising better balance between speed and accuracy.