Post‐Constrained Conditional Sedimentary Facies Simulation From a Single Training Image Using a Concurrent Multi‐Stage Generative Adversarial Network

Conditional sedimentary facies modeling enables accurate characterization of the spatial architecture of sedimentary facies, providing crucial support for the fine evaluation and efficient development of natural resources. However, traditional conditional generative adversarial networks (CGANs) often require extensive training data and trade‐offs in conditional loss, making it difficult to maintain both diversity and geological consistency in data‐scarce geological scenarios. To address this issue, this study proposes a conditional simulation method integrating a training network and a post‐constraint network (CMS‐PCSinGAN). The training network adopts a concurrent multi‐stage SinGAN architecture based on a single training image to accelerate model convergence. The post‐constraint network progressively incorporates sparse well facies data through multi‐scale image editing to enhance the geological consistency of the generated realizations. To validate the effectiveness of the proposed method, experiments were conducted on multiple fluvial sedimentary facies models, and the results were systematically evaluated using facies proportion analysis, variogram analysis, and connectivity analysis. The results demonstrate that the CMS‐PCSinGAN method can satisfy sparse conditional constraints while maintaining strong diversity and geological consistency. Compared with commonly used multiple‐point statistics (MPS) methods, the proposed approach shows significant advantages in both applicability and reproduction of statistical characteristics.