Embedded Discrete Fracture Modeling With Artificial Intelligence in Permian Basin

You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.

To ensure continued access to JPT's content, please Sign In, JOIN SPE, or Subscribe to JPT

Full-field models using unstructured grids can capture detailed geometric information such as fracture distribution. However, these are computationally expensive and often numerically unstable because of convergence issues. In the complete paper, the authors investigated embedded discrete fracture modeling (EDFM) using artificial intelligence (AI) to overcome challenges associated with unstructured modeling.

Introduction

It has been proved that EDFM enables flexible fracture geometry because the fracture domain is relatively independent of the matrix regions. EDFM has been widely accepted recently because of its simplicity and computational efficiency. The authors’ studies applied EDFM with AI optimization for fracture-network representation.

The growth of computational power and the availability of large quantities of data have led to the widespread promotion of applications of AI technologies. AI has been used in the oil and gas industry in the areas of production optimization, operating-cost reduction, and efficiency improvement. Currently, almost all AI technologies are still limited to executing specific tasks, also known as specialized AI, an application that lacks the capability of generalized AI in adaptive learning. However, specialized AI still has advantages of high speed, superior consistency and unmatchable repeatability over human intelligence. In the complete paper, application of AI technology in processing the fracture network for successive simulation is described.

Methodology

EDFM. This method has drawn considerable attention because of its efficiency. EDFM minimizes the local grid resolution while preserving flow behavior between fracture and matrix by maintaining the original fracture orientation and distribution. The EDFM method is composed of two major elements: matrix and fracture. A long fracture is subdivided into small segments at the interconnecting points. A simple material-balance equation can show the flow behaviors at the cross sections.

The fractures and matrix are modeled as two different domains. As described previously, fluid communication between fractures and the matrix uses mass-balance equations. Nonneighbor connection (NNC) must be calculated explicitly for a reservoir simulator.

EDFM With Mangrove. Mangrove, an engineered stimulation design package, is a hydraulic-fracture simulator that links reservoir characterization and simulation and helps optimize completion designs in unconventional reservoirs. It has been of considerable use in unconventional-field development and has provided guidance for well-landing, completion-design, and well-/fracture-optimization decisions. However, the package creates unstructured simulation grids that can lead to convergence issues, and is also computationally expensive.

In this study, fracture geometry and properties (conductivity, aperture, and porosity) are exported directly from Mangrove models through the Petrel platform. With the fracture information, EDFM models are created with respect to the true fracture distribution and properties. Fracture segments may or may not penetrate the entire matrix domain in different locations.

After the fracture geometry and properties are collected from the package, the EDFM grids are created. The fracture geometry is honored when transferred to EDFM grids.

AI Optimization. To address various challenges of the fracture structure created by Mangrove, such as difficulty in convergence and high computational expense, petroleum engineers would preprocess the fracture structure before reservoir modeling and simulation. The preprocessing steps include, but are not limited to, identification of the major fractures, selection of fractures that were most representative of the overall network, and removal of useless fracture patches such as closed fractures or those without proppant. This manual process is onerous, time-consuming, and not reproducible because of the different criteria adopted by different petroleum engineers.

AI can serve as the data-processing agent to complete the task in an efficient, accurate, and repeatable manner. An expert system was created to simulate petroleum engineers’ tasks of identification, selection, and removal. The selection criteria were developed by use of engineers’ input and took into consideration constraints such as physics, energy, and fracture-growth mechanisms.

The current approach could handle any type of fracture system but, like any AI technology, lacked adaptive learning capability. One obvious limitation is the inability to handle abnormal situations, such as an isolated fracture that petroleum engineers could readily identify and handle.

Mangrove/AI/EDFM. Mangrove discretizes the entire fracture into small segments that are assigned to different properties. The AI tool used by the ­authors is intended to maximize the possible fracture surface while minimizing the energy cost.

All fracture segments created in the Petrel/Mangrove package are exported for EDFM gridding purposes. The fracture properties, including porosity, permeability (conductivity), and aperture, are preserved and transferred for data preprocessing. As seen in Fig. 1, the exported structures are very complex, and the fractures are crosslinked from stage to stage. After the AI optimization, the small fractures are eliminated from the model and fracture connections are also reduced significantly. Removing the small fractures from the models has one obvious advantage in regard to simulation: The total simulation cell count is decreased accordingly. However, the elimination of small fractures could also lead to a great reduction in the fracture surface and connections that would have a significant contribution to the overall reservoir production. Therefore, in the authors’ simulation studies, for the AI optimized cases, engineering-matching methods are required to compensate for the loss of detailed fractures and their connections.

Fig. 1—Fracture optimization/simplification. Left: original; right: optimized.


 

Model Validation and Results

The proposed approach is validated by comparing the simulation performance of the EDFM case with the Mangrove-created unstructured simulation cases. Three different field-scale cases are presented to validate the effectiveness and efficiency of the proposed EDFM method:

  • A single horizontal well with multistage fractures applying the AI optimization
  • A single horizontal well with multistage fractures without the AI optimization
  • Three horizontal wells with multistage fractures without the AI optimization

For the first two cases, a reservoir model is prepared to simulate the shale oil case. The authors tested the EDFM effectiveness and efficiency with and without AI assistance. For the three-well case, the effectiveness of EDFM in the Permian Basin was validated and an effort was made to capture the well interference. To simplify the problem, the matrix is defined as a layer-cake model, which has different permeabilities and porosities from layer to layer. The fracture geometry and properties are exported from Mangrove without any tuning.

Conclusion

In this work, the authors applied different combinations of work flows, including Mangrove/EDFM and Mangrove/AI/EDFM, to investigate the potential, effectiveness, and efficiency of EDFM in further field development. EDFM has proved to be reliable and efficient for fracture-model simulation in the shale play. The authors have tested different scenarios with and without AI for further efficiency improvement. One of the major advantages of using EDFM with the optimized structure is that sensitivity analysis is applicable for better decision making in field planning, such as in well-landing, fracturing, and well-­spacing studies. Even with the nonoptimized original fracture network, EDFM has shown a major improvement in computational efficiency compared with the unstructured simulation, though many different factors could contribute to this improvement. EDFM provides a way of representing the fractures explicitly inside a structured matrix. This study combines EDFM and AI techniques to perform numerous full-field simulation runs, making uncertainty analysis affordable for better planning.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 187202, “Field Study: Embedded Discrete Fracture Modeling With Artificial Intelligence in Permian Basin for Shale Formation,” by Song Du, Baosheng Liang, and Lin Yuanbo, Chevron, prepared for the 2017 SPE Annual Technical Conference and Exhibition, San Antonio, Texas, USA, 9–11 October. The paper has not been peer reviewed.

Embedded Discrete Fracture Modeling With Artificial Intelligence in Permian Basin

01 May 2018

Volume: 70 | Issue: 5

STAY CONNECTED

Don't miss out on the latest technology delivered to your email weekly.  Sign up for the JPT newsletter.  If you are not logged in, you will receive a confirmation email that you will need to click on to confirm you want to receive the newsletter.