GNGTS 2023 - Atti del 41° Convegno Nazionale

Session 3.3 ______ ___ GNGTS 2023 Conclusion We exploit the Hidden Layers (HL) of LSTM networks as Deep Attributes (DA) of reflection seismic data sets. The results obtained from synthetic and field data show that the new method is able to manage even complex geometries highlighting not only single seismic reflectors (i.e., features with high spatial frequency) but, even more important, the main geological and geophysical features related for instance to the low spatial frequency seismic velocity trend. The stability tests performed demonstrate the high affordability of the procedure that can be trained on a single data set (or just on a portion) and then applied to larger seismic volumes. The application of DA to a 4-D data set collected to monitor a controlled CO 2 injection in an underground gas storage further demonstrate the stability and the repetitiveness of the methodology and in turn its full applicability for monitoring purposes. Future research will be directed to infer specific correlations between DA and single or integrated physical parameters which would make the DA a new possible strategy for the quantitative subsurface petrophysical characterization at different scale and resolution levels. Acknowledgements This research was partially supported by PNRA projects IPECA (PNRA18_00186) and CRIOVEG (PNRA18_00288) and by the project “Dipartimento di Eccellenza" of the Department of Mathematics and Geosciences of the University of Trieste. We gratefully acknowledge the support of Halliburton Landmark and Shearwater through the University Grants Program (UGP). References Anstey N., Bahorich M.S., Bridges S. R., Farmer S.L., et al., 2007, Overview of Seismic Attributes, Geophysical Developments Series: 1-24, https://doi.org/10.1190/1.9781560801900.ch1. Chopra S., Marfurt K.J., 2005, Seismic attributes — A historical perspective, Geophysics, 70, 5, 3SO-28SO, https://doi.org/10.1190/1.2098670. Hochreiter, S., and Schmidhuber, J., 1997, Long short-term memory, Neural computation, 9(8), 1735-1780, https://doi:10.1162/neco.1997.9.8.1735.