Predictive geology, with the aid of machine learning (ML) and geophysical inversion tools, has witnessed significant achievements in recent times. The present-day ensemble learning algorithms are capable of generating virtual boreholes with high accuracy using spatial data and stratigraphic sequencing data. Moreover, unsupervised machine learning is also capable of geological differentiation, i.e., differentiation of unique geological formations in greenfield areas with limited prior knowledge.
This is a cost-effective option compared to conventional resource definition drilling, which is not only expensive but also harmful to the environment. Though the present-day predictive models are highly efficient, they are still limited by their dependency on actual data.
For instance, hybrid models of theory-guided data science have been found to improve reservoir characterization by using simulated ‘pseudowells’ to train neural networks, which are statistically constrained to actual well data from the study area (Downton et al., 2020).
Similarly, ML models are now capable of accurately predicting rock porosities using real-time drilling parameters, which is a cost-effective and time-saving option compared to conventional laboratory analysis. However, it is mandatory to have a dataset with porosity values obtained from core analysis for training and validating ML models (Ouladmansour et al., 2023).
The term ‘hybrid intelligence’ is another term for such a collaborative approach between human geologists and machine learning models using geophysical data to construct probabilistic models that maximize future well locations and densities (Ball & O’Connor, 2021). This ensures that every well drilled maximizes the amount of information gained to improve the geological model.
In conclusion, predictive geology is not a replacement for drilling but rather a supplementary tool to conventional drilling and sampling, which is essential for generating continuous calibration data for predictions and hence plays a part in a more efficient exploration strategy.
References
Ball, A., & O’Connor, L. (2021). Geologist in the Loop: A Hybrid Intelligence Model for Identifying Geological Boundaries from Augmented Ground Penetrating Radar. Geosciences, 11(7), 284. https://doi.org/10.3390/geosciences11070284
Downton, J., Collet, O., Hampson, D., & Colwell, T. (2020). Theory-guided data science-based reservoir prediction of a North Sea oil field. The Leading Edge, 39, 742–750. https://doi.org/10.1190/tle39100742.1
Ouladmansour, A., Ameur-Zaimeche, O., Kechiched, R., Heddam, S., & Wood, D. A. (2023). Integrating drilling parameters and machine learning tools to improve real-time porosity prediction of multi-zone reservoirs. Case study: Rhourd Chegga oilfield, Algeria. Geoenergy Science and Engineering, 223, 211511. https://doi.org/10.1016/j.geoen.2023.211511
