Operations and execution

A machine learning framework for rapid forecasting and history matching in unconventional reservoirs

“…These workflows employ detailed physics models, referred to as high-fidelity models in this paper, to perform simulations that are expensive to run. For example, it can take several days to months to run reservoir-scale model simulations 2 – 4 with degrees-of-freedom of the order of hundreds of millions on state-of-the-art high-performance computers (HPC). In other words, these current physics-based approaches are not amenable to real-time decisions even for conventional reservoirs that are dominated by porous flow.…”

“…Unconventional reservoirs pose an even greater challenge for real-time forecasting: the physics of fluid flow is a complex combination of processes in micropores (< 2 nm) and mesopores (2–50 nm) and in comparatively larger fractures 5 . Unconventional reservoirs typically have fractured effective porosity in the range of 0.04–0.08 and fractured permeability of the order of nanodarcies ( – ) 2 , 4 . Note that the porosity and permeability of shale are an order of magnitude less than fractured shale 6 .…”

“…However, the strategies and workflows used in conventional reservoirs are either inapplicable, or prohibitive in case of unconventionals. Due to the long horizontal laterals and fewer wells drilled, there is insufficient reservoir characterization data to inform high-fidelity physics models 4 , 11 . Moreover, history matching with high-fidelity models takes time during the early phase of a well’s production when the majority of hydrocarbons are produced, and the inherent rapid initial decline in production rates and the unpredictability of the yield in unconventionals suggests that the window of opportunity to exploit the reservoir is limited.…”

“…As a result, workflows with high-fidelity simulations are not ideal for usage in comprehensive uncertainty quantification studies that require 1000s of forward model (forecasting) runs. The unpredictability is suspected to be due to the poorly understood nature of the physics of the drawdown process, which is exacerbated by the impact of resource development processes themselves (e.g., slow drawdown or fast drawdown) 2 , 4 . Thus innovative extraction strategies and advanced workflows could significantly improve the hydrocarbon recovery efficiency and long-term economics 12 – 14 .…”

“…Current workflows for unconventional reservoirs are predominantly based on production decline curve analysis and its extensions 15 – 17 , data-driven machine-learning (ML) approaches 11 , 18 – 26 , and/or extension of physics-based conventional reservoir workflows 2 – 4 , 27 – 30 . Decline curve analysis provides empirical models to forecast production data based on the past production history.…”