Efficient and cost-effective unconventional oil and gas (UOG) recovery depends critically on the knowledge of primary factors controlling the reservoir producing behaviors, as well as on well completion strategies. Currently the completion design of UOG wells is often dominated by geometry-based approaches, neglecting the impact of spatial heterogeneity of reservoir properties. The primary goal of this work is to identify geological factors and well completion strategies important to production using systematic Design of Experiment (DoE) methodologies, and then train a data-driven, machine-learning (ML) proxy model to expedite optimization of well completion. The results are demonstrated for applications in Permian Basin.

A set of Permian Basin wells are selected to provide a wide spectrum of geological, geomechanical and completion features existing in the basin. For each well, process-level modeling is performed in commercial hydraulic fracturing (HF) and reservoir simulators. For each well, the HF model is calibrated against historical well production data, by adjusting hydraulic fracture structure and reservoir properties. Using DoE methodologies, we evaluate a large number of completion strategies, in conjunction to different geological and geomechanical conditions. The effect of different decision variables on HF completion efficiency and production are examined, including the type of proppant carrier fluid (slick water and crosslinker), and proppant types (e.g., ceramic, curved resin, sizes, concentration, size, etc.). The results are used to develop an ML-based proxy model, which can be used to make rapid design of well completion strategies for future development, without requiring running time-consuming, full-scale reservoir simulations.

Simulation results of this work show the well completion implementation for many of the selected wells is far from optimal. History matching helps to establish the input-production relationships for each well, which provides a base model for sensitivity runs. The well set is divided into two parts. One part is used to develop an ML proxy model, and the rest is used for testing. Proxy modeling results suggest that the machine learning model can learn the complex reservoir input-output relations well, providing a data-driven tool for rapid well completion design and field production evaluation.

The complex geology, coupled with dynamic fracture stimulation and reservoir production processes are often not well investigated and represented in current practice, due to the lack of tools and representative data. As a result, sub-optimal production prevails, which erodes the HF economics. Previous works either mainly focus on mechanism simulation study or over-emphasize the original raw data but neglect the complex geology and physics underlying fracturing and reservoir production processes. The contribution of this study is we provide an integrated solution for well completion design based on real production data, fracturing simulation and reservoir simulation from different representative locations in Permian Basin. The physics-based proxy model can be used for future well designs by taking account into the effects of a large number of geological, geomechanical, and production factors.

This content is only available via PDF.
You can access this article if you purchase or spend a download.