Speaker
Description
Reservoir history matching aims to infer key subsurface parameter fields, such as permeability, from dynamic production data, and it provides an essential basis for reservoir simulation and development decision-making. Early history matching mainly relied on expert knowledge and repeated trial-and-error adjustments, which were labor-intensive and highly subjective. It later evolved into optimization/assimilation-based iterative methods, represented by ESMDA, where parameters are calibrated through multiple updates; however, these methods still suffer from high computational cost, long iteration cycles, and limited efficiency in complex nonlinear settings. In recent years, with the rapid development of deep learning and generative modeling, end-to-end history matching frameworks that directly map production data to permeability fields have emerged as a more promising direction, significantly improving inference efficiency while reducing human intervention. Building on this, we propose an end-to-end history matching method based on a Visual AutoRegressive (VAR) generative model. The permeability field is represented as a generatable discrete sequence, and an autoregressive mechanism is used to learn the conditional distribution from production data to parameter fields, enabling fast generation and inversion. Comparative experiments show that, compared with the conventional ESMDA approach, the proposed method substantially accelerates inference while maintaining matching accuracy, thereby reducing overall computational overhead. Furthermore, we incorporate well-point constraints into the end-to-end generation process by explicitly injecting known and reliable information at well locations, which narrows the feasible solution space and reduces inversion uncertainty, improving the stability and credibility of the generated permeability fields.
| Country | China |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
