Three-dimensional basin basement boundary inversion of gravity anomalies based on EfficientNetV2 network

Geophysical exploration is a vital technical approach, offering crucial data support for oil and gas exploration, representing a significant aspect of the industry. Oil and gas deposits are commonly found in basins or certain folds. Traditionally, seismic exploration is employed to outline basin interfaces for oil and gas detection. However, conventional gravity regularization inversion is constrained by stability and resolution issues, impeding precise characterization of sudden density interfaces. To address this, we construct layered geological models to gather gravity anomaly data, subsequently applying artificial intelligence for data feature extraction and geological model inversion. This process serves as a preliminary step for seismic exploration. Our findings indicate that the inversion prediction method utilizing EfficientNetV2 more effectively predicts the base boundary and density variations of 3D basin models. This method surpasses ResUnet++ and DNN in terms of accuracy, training speed, and reduced training parameters, demonstrating broad application potential. Beyond enhancing basin gravity exploration’s accuracy and efficiency, this approach offers a novel insight into basin structure and evolution.