DISTINGUISH Workflow: A New Paradigm of Dynamic Well Placement Using Generative Machine Learning

TL;DR

DISTINGUISH presents an integrated, real-time geosteering framework that fuses GAN-based geomodeling with ensemble data assimilation and a naive-optimistic ADP optimizer to continuously refine subsurface uncertainty and guide drilling decisions. By offline training GANs to generate geologically plausible realizations and a fast forward model for LWD responses, the workflow updates predictions online via EnKF and selects trajectory steps through dynamic programming. In a synthetic benchmark, DISTINGUISH demonstrated adaptive steering toward high-value sand bodies and notable uncertainty reduction, while highlighting artifacts and the need for further methodological refinements. The approach holds promise for automated, data-driven geosteering and real-time reservoir characterization, with open-source code planned to accelerate community adoption and development.

Abstract

The real-time process of directional changes while drilling, known as geosteering, is crucial for hydrocarbon extraction and emerging directional drilling applications such as geothermal energy, civil infrastructure, and CO2 storage. The geo-energy industry seeks an automatic geosteering workflow that continually updates the subsurface uncertainties and captures the latest geological understanding given the most recent observations in real-time. We propose "DISTINGUISH": a real-time, AI-driven workflow designed to transform geosteering by integrating Generative Adversarial Networks (GANs) for geological parameterization, ensemble methods for model updating, and global discrete dynamic programming (DDP) optimization for complex decision-making during directional drilling operations. The DISTINGUISH framework relies on offline training of a GAN model to reproduce relevant geology realizations and a Forward Neural Network (FNN) to model Logging-While-Drilling (LWD) tools' response for a given geomodel. This paper introduces a first-of-its-kind workflow that progressively reduces GAN-geomodel uncertainty around and ahead of the drilling bit and adjusts the well plan accordingly. The workflow automatically integrates real-time LWD data with a DDP-based decision support system, enhancing predictive models of geology ahead of drilling and leading to better steering decisions. We present a simple yet representative benchmark case and document the performance target achieved by the DISTINGUISH workflow prototype. This benchmark will be a foundation for future methodological advancements and workflow refinements.

Figures (7)

• Figure 1: Overview of the DISTINGUISH workflow: (a) the online phase, and (b) a detailed schematic of the "full mapping sequence" highlighting its inputs and outputs.
• Figure 2: Resistivity of the synthetic truth (in Ohm-m). The line shows a possible well trajectory with stars in the measurement locations; the arrows show the tool sensitivity of up to three boundaries in an up or down direction. The measurement error is set to 10% relative noise.
• Figure 3: The prior model at the start of the operation with a star indicating the kick-off drilling position. The prior indicates approximately 40% probability of a sand body in the horizontal direction ahead of the bit. The image shows the probability of sand (either channel or crevasse) for every point based on the ensemble average. The white outline indicates the synthetic truth unknown to the DSS.
• Figure 4: The weighted reward image for the synthetic truth, showing the relative value for reaching each of the cells. The cell values are computed according to Equation (\ref{['eq:reward']}).
• Figure 5: Illustration of the DISTINGUISH decision support system (DSS) applied to a synthetic scenario. This figure details the step-by-step decision recommendations provided by the DSS, indicated by the proposed-direction dashed line and advancing the bit in the next sub-figures. The "fan" of thin dashed trajectories, which start from the recommended decision, indicates path possibilities optimal for individual realizations in the ensemble. The trajectory paths in the fan are perturbed: thicker line agglomerates correspond to more trajectories targeting each area. The scenario begins with an initial geomodel uncertainty, which gradually decreases as EM measurements are collected during drilling (shown by stars). The background images show the probability of sand (channel or crevasse) after data assimilation at the current step. The outline of the true model is provided as a reference but is unknown to the DSS.