3D Machine Learning Inversion
Multi-Physics Technologies offers 3D Machine Learning Inversion (3DMLI) to predict subsurface geology for reservoir characterization studies. We accomplish this by integrating seismic and well data to predict high frequency acoustic, elastic and petrophysical properties in three dimensions. Our technology is backed by advanced machine and deep learning algorithms using ensemble learning and neural networks.
Our 3DMLI workflow may be used as an alternative to the conventional seismic inversion process that has been around for decades. The most important limiting factor to conventional inversion is that the input seismic frequency content mostly governs the resolution power of 3D characterization models. Additionally, this workflow must be carried out in the time domain and requires multiple steps to predict petrophysical and reservoir properties. On the contrary, our workflow is a one step process for predicting most measured log properties in three dimensions, with bandwidth beyond seismic that can be carried out in the depth or time domain (Figure 1).
Figure 1: Comparison of conventional QI and 3D Machine Learning Inversion workflows.
Figure 2 shows the required inputs for 3D Machine Learning Inversion and how they compare to conventional prestack seismic inversion. The main differences are that High-Def does not require the estimation of wavelets and multiple properties are not inverted for simultaneously. Additionally, the low frequency model (LFM) is no longer generated solely from well data.
The input data types include prestack 3D, multi-component, 4D and azimuthal seismic data. The Hybrid Model is generated using extrapolation/interpolation methods, variograms or other statistical methods. The model requires surface grids, seismic dip and/or azimuth. Optional input includes velocities from seismic processing.
Figure 2: Required inputs for conventional prestack seismic inversion (left) and 3D Machine Learning Inversion (right).
In machine learning studies, having more than one well is essential since a well used to build the ensemble machine learning model will usually result in a near perfect match with the input data. For this reason, blind well tests are an effective way to determine how trustworthy the modeled property will be. Figure 3 shows how blind wells play a role in the quality control of a characterization study with two wells.
Figure 3: Generalized workflow using 2 wells.
Case Studies
- Garcia Leiceaga, G, Balch, R., and El-kaseeh, G., (2021), Subsurface Characterization Using Ensemble Machine Learning, Offshore Technology Conference, Houston, USA
- Garcia Leiceaga, G., and Rimaila, K., (2021), Resolving Sedimentary Features in a Lower Miocene Clastic Reservoir, Jeanerette Field, IMAGE – AAPG and SEG Annual Convention & Exhibition, Denver, USA
References and Further Reading
- Aki, K. and Richards, P.G., 1980, Quantitative Seismology – Theory and Methods: W.H. Freeman and Company.
- Breiman, L., Friedman, J., Stone, C. J. and Olshen, R.A., 1984, Classification and Regression Trees, Taylor and Francis.
- Breiman, L., 1996, Bagging predictors, Machine Learning, 26(2), 123–140.
- Breiman, L., 2001, Random Forests, Machine Learning, 45, 5–32.
- Bach, R., Will, R., El-Kaseeh, G., Grigg, R., Hutton, A. and Czoski, P., 2015, Integrating Multi-Scale Seismic Measurements for EOR/CCUS, 85th Ann. Internat. Mtg: Soc. Of Expl. Geophys., 2837-2841.
- Draper, N. R. and Smith, H., 1998, Applied Regression Analysis, Wiley-Interscience
- Hastie, T., Tibshirani, R. and Friedman, J., 2009, The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Springer, New York, NY.
- Ho, T. K., Random Decision Forests, 1995, Proceedings of 3rd International Conference on Document Analysis and Recognition, vol. 1, 278–282.
- Ho, T. K., 1998, The Random Subspace Method for Constructing Decision Trees, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, no. 8, 832-844.
- Kwok, S. W. and Carter, C., 1988, Proc. Fourth Int. Conf. Uncertainty Artificial Intell., 327-338.
- Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M. and Duchesnay, E., 2011, Scikit-learn: Machine Learning in Python, JMLR, 12, 2825-2830.
- Shuey, R.T., 1985, A simplification of the Zoeppritz equations, Geophysics, 50, 609-614.
- Zoeppritz, K., 1919, VIIb. Über Reflexion und Durchgang seismischer Wellen durch Unstetigkeitsflächen. [VIIb. On reflection and transmission of seismic waves by surfaces of discontinuity], Nachrichten von der Königlichen Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-physikalische Klasse, 66–84.