Geology-informed deep learning for basin restoration and deep-time source-to-sink reconstruction—a case study in the Western Bonan Low Uplift, Bohai Bay Basin

Understanding the sedimentary framework and reconstructing deep-time source-to-sink (S2S) systems in rift basins are critical for predicting reservoir distribution and reducing exploration risk in structurally complex offshore settings. We present a geology-informed Bayesian deep learning framework that explicitly encodes geological priors within graph neural networks, variational autoencoders and attention-based architectures, and combines these with probabilistic sampling to generate uncertain