(589c) Production Estimation and Well Classification for Hydraulically Fractured Horizontal Wells: A Data-Driven Model-Based Approach

Hydraulic fracturing is essential for oil and gas production from unconventional low-permeability reservoirs. Fracturing of horizontal wells ushered in the era of economic production of natural gas from unconventional reservoirs. A key challenge in hydraulic fracturing is how to generate optimal fracturing jobs, i.e. to determine the values of basic variables related to fracture construction and properties, such as spacing and dimensions. Decisions on such variables would optimize a related criterion, e.g. production, recovery, or NPV. Ideally, it would be desirable to understand the effect of several decision variables as well as reservoir-related parameters on a chosen objective. To this end, fundamental equations can be used that describe phenomena related to hydraulic fracturing and subsequent production from a fractured reservoir. This is currently the prevailing approach in industry. However, there is by now a significant body of data on production from unconventional fractured wells. Therefore, an opportunity exists to explore how such data can be used to develop methodologies that efficiently suggest optimal fracturing designs through the use of data-driven predictive models. Such models would capture the combined effect of model input variables, namely fracturing-job variables and reservoir properties, on a chosen objective. The purpose of this presentation is to explore the possibility of developing a data-driven modeling approach to hydraulic fracturing, suitable for design purposes.

There are numerous techniques for developing data-driven models. They all combine art with science. An approach that is first explored in this research is PLS (Partial Least Squares) to build a model connecting various input variables (such reservoir properties, wellbore geometry, fracture construction process, and others) with production. Given that production varies with time, a second target is the development of a dynamic model, capturing the effect of decision variables on production over time. Sensitivity analysis of decision variables is carried out to identify variables with high impact. The VIP (variable importance in projection) statistic is explored to that end.

The data-driven models developed will be useful for design purposes and what-if analyses. They can also provide insight into the variables that have strong impact on fractured horizontal well performance. Therefore, the proposed models are envisioned to offer initial decision support for any fracturing operation. In addition, they may will also be able to indicate parameters that should be improved/modified to get desired results in existing wells.

Future refinements of the proposed approach include the development of nonlinear models from a variety of well established multivariate statistical methods.