State Space Modeling of Drilling Data to Automatically Infer Lithology

Abstract Surface drilling data contains a wealth of information related to rock petrophysical and mechanical properties that can be harnessed for near real-time reservoir characterization. This work provides a look-ahead for the bit to anticipate changes in lithology rapidly and with great fidelity. The forecasting discussed in this paper is based on state-space modeling that is probabilistic and provides a distribution over possible lithologies rather than a deterministic outcome. This allows for the optimization of drilling operations at the surface to simultaneously maximize rate of penetration (ROP) and minimize bit wear. State-space models are powerful in sequentially modeling dynamic systems that are partially observable and evolve over time. In our study, the states represent the underlying petrophysical properties (or lithologies) which are rarely observed in real-time. The observations are represented by real-time surface drilling data. Our framework predicts the future sequence of lithologies, taking into account both the past sequence of observations and the inferred transition dynamics between lithologies. We evaluate our state space model on two field-case studies: The first is the well-known Volve Field data provided by Equinor and the second is an unconventional play in Oklahoma. In both case studies, we demonstrate how the state space models can use surface drilling data to infer changes in lithology in the subsurface. From these transitions in lithology, we can derive a probabilistic model to forecast the likely sequence of states (or lithologies) the bit is likely to see soon. Remarkably, we achieve high accuracies in the near-term; however, these accuracies diminish for longer-range predictions. Nevertheless, for the drilling crew, the information we provide by looking ahead of the bit is valuable information to anticipate and prepare for changes to drilling strategy, mud weight, weight-on-bit and other operational variables. State-space models are a relatively underexplored tool for modeling drilling systems that evolve dynamically over time. While autonomous drillers have reduced the burden on drilling crews, anticipating lithology changes can provide valuable information for geosteering applications, to optimize rate-of-penetration and to prevent excessive bit wear.