Third EAGE Conference on Pre Salt Reservoirs

Using data science and machine learning techniques, such as image segmentation, and object classification, we have been able to predict; lithology from core images, grain type and microfacies as well as generate pore property statistics from thin section photomicrographs and generate auto picks and facies from borehole image logs. By applying these data science techniques to traditional geoscience data, we are able to come up with innovative solutions which generate large volumes of data with quantified degrees of confidence and a reduction in interpreter bias. The results of these approaches demonstrated an excellent correlation with independent datasets including: x-ray diffraction spectroscopy, helium porosity, total organic carbon and both core and wireline gamma; indicating these results represent a reliable and rapid proxy for many geological and petrophysical parameters. In using data analytical techniques to gain more quantitative data from images, including legacy images, we can quickly and accurately gain insight and understand trends within individual wells, fields, basins or regions.

The geological reservoir characterization is an essential task for planning exploratory targets and production strategies. In this context, to know the rock textures and structures is the link to understand pore network properties in the geological formations. Therefore, the reservoir facies analysis is fundamental to define the better permo-porous reservoir intervals. Thus, this work presents the main results obtained from the use of Deep Learning models in the automatic facies classification applied into the pre-salt well cores of Santos Basin. The development of this innovative methodology was based on training of neural network algorithms with high definition well core image to predict facies automatically in untrained analogue well cores. The results achieved a satisfactory overall accuracy around 60%; hit values higher than 70% were found mainly in the most representative carbonate reservoir facies as coquinas, reworked limestones represented by rudstones and grainstones, and in situ facies such spherulitites and shrub limestones. This new methodology, based on the artificial intelligence, has potential to become a powerful tool to be used in the large scale for the Brazilian pre-salt carbonate reservoir characterization.

The calibration of petroelastic models is an important step to generate exploratory scenarios and to estimate accurate petrophysical properties from elastic inversion and amplitude versus offset results. In this work we discuss the successful application of three rock physics models to pre-salt rocks from Santos Basin: Stiff Sand model (SSM), Vernik-Kachanov model (VKM) and Differential Effective Media (DEM). The results showed that any of these three models can be used in pre-salt carbonate rocks to predict elastic properties, and the choice of a particular model should be made based on the available parameters and on the application, since each model may be more suitable for specific geological feature, like diagenesis effects studies.

In this study, the examples of pore throat size distribution (PSD) clusters/classes derived from Mercury injection capillary pressure (MICP) data as rock type identifier are shown on different carbonate reservoirs as the first step in a systematic rock typing workflow. The identified rock types are also called petrophysical rock types (PRTs) by Thomeer (1983) possessing a cluster or continuum in the parameter space related to petrophysical attributes such entry pressure corresponding to the largest pore throats, the distribution of the pore throats and total amount of porosity which can be extracted from MICP analysis as well as permeability.

More than 400 MICP data were analysed, where the clusters of the derived PSDs are compared against porosity/permeability grouping for the corresponding core plugs. In this study, it is shown how PSDs or Thomeer hyperbola’s parameters grouping can be useful to identify the number of rock types/PRTs for carbonate reservoir characterisation. A good agreement between porosity/permeability grouping and clustering of either PSDs or Thomeer hyperbola’s parameters are also observed.