InterPore2026

Multi‑scale AI‑enabled production forecasting for shale gas: integrating digital rock physics, geo‑engineering descriptors and field time‑series

20 May 2026, 15:35

1h 30m

Poster Presentation (MS15) Machine Learning in Porous Media Poster

Speaker

Runshi Huo (PetroChina Research Institute of Petroleum Exploration and Development)

Description

Accurate forecasting of well production is critical for managing shale gas development, yet remains challenging because of multiscale heterogeneity, strong geological–engineering coupling and complex flow regimes in ultra tight, multi porosity media. Here we develop a multi scale, AI enabled workflow that integrates digital rock physics, geological and engineering descriptors, and field production time series to predict well level production dynamics in hydraulically fractured horizontal wells. High resolution digital core images are processed with deep learning–based image analysis to efficiently extract pore and throat scale properties, including porosity, permeability and pore network connectivity at micro to nano scale. A supervised upscaling model then maps these digital rock derived features onto horizontal well segments, yielding digitally constrained static reservoir properties for the target intervals. In parallel, 24 macroscopic geological and engineering parameters are selected to capture large scale controls on flow. The digital rock descriptors and macro scale geo engineering parameters are jointly fused with field production time series within a hybrid deep learning framework, in which multi scale static features condition the temporal encoder to introduce physics informed constraints into data driven forecasting. Application to a shale gas field case demonstrates that the proposed method outperforms conventional decline curve analysis and purely data driven models in predicting production dynamics, delivering higher accuracy and more reliable guidance for production management. The results highlight that digital rock physics can serve not only for fine scale petrophysical characterization, but also as high dimensional, high information static descriptors for production forecasting, providing a practical pathway to bridge pore scale imaging with field scale shale gas development and optimized production strategies.