Strategic Geosteering Workflow with Uncertainty Quantification and Deep Learning: A Case Study on the Goliat Field

TL;DR

This paper presents a real-time geosteering workflow that integrates a DNN surrogate of EDAR forward modeling with FlexIES, an ensemble-based Bayesian smoother designed to account for unknown model errors. The offline phase builds geologically-relevant priors and trains the DNN on physics-consistent data, while the online phase performs rapid assimilation of EDAR measurements to produce probabilistic layer boundaries and resistivities under anisotropic conditions. Case study results from the Goliat field demonstrate that the posterior median matches a proprietary deterministic inversion, and FlexIES provides meaningful uncertainty quantification and automatic handling of multi-modality and model-error effects. The approach supports robust, real-time decision-making in challenging subsurface environments and offers avenues for extending to more complex priors and additional measurement types.

Abstract

The real-time interpretation of the logging-while-drilling data allows us to estimate the positions and properties of the geological layers in an anisotropic subsurface environment. Robust real-time estimations capturing uncertainty can be very useful for efficient geosteering operations. However, the model errors in the prior conceptual geological models and forward simulation of the measurements can be significant factors in the unreliable estimations of the profiles of the geological layers. The model errors are specifically pronounced when using a deep-neural-network (DNN) approximation which we use to accelerate and parallelize the simulation of the measurements. This paper presents a practical workflow consisting of offline and online phases. The offline phase includes DNN training and building of an uncertain prior near-well geo-model. The online phase uses the flexible iterative ensemble smoother (FlexIES) to perform real-time assimilation of extra-deep electromagnetic data accounting for the model errors in the approximate DNN model. We demonstrate the proposed workflow on a case study for a historic well in the Goliat Field (Barents Sea). The median of our probabilistic estimation is on-par with proprietary inversion despite the approximate DNN model and regardless of the number of layers in the chosen prior. By estimating the model errors, FlexIES automatically quantifies the uncertainty in the layers' boundaries and resistivities, which is not standard for proprietary inversion.

Figures (26)

• Figure 1: The full workflow for preparation and real-time data assimilation during drilling.
• Figure 2: The schematics of the DNN architecture. The DNN is composed of five consecutive fully-connected layer pairs ($Ai$, $Bi$), each having ReLU activation. The outputs of the layer pairs are summed together, forming structure similar to residual blocks, denoted $\beta_i$. The pairs are followed by the linear output layer. The numbers in each box are the sizes trainable parameters, and the numbers on the arrows are the sizes of inputs / outputs of layers.
• Figure 3: a. The reference vendor-provided inversion described in larsen2015extra. b. The case study section of the well for this paper. Dots show the well position in the middle of the inversion intervals along the well trajectory.
• Figure 4: Schematics of 1D input from a training dataset for the DNN forward model.
• Figure 5: Schematic diagram of FlexIES algorithm for real-time inversion of EM measurements by utilizing DNN as a forward model.