Wykorzystanie technik sztucznych sieci neuronowych do predykcji wybranych parametrów jako uzupełnienia zbioru danych wejściowych w konstrukcji modeli parametrycznych 3D

Abstract

The article presents the possibilities of using artificial neural networks for parametric prediction in borehole profiles, the application of which supplemented the set of information in all boreholes located within the analyzed area. The approach presented in the article will be used when there is no possibility of specialized interpretation of the drilling geophysics curves, supplementing the missing data. The set of data used in the study included solutions in the profiles of 10 boreholes, four of which were characterized by the availability of the full data set analyzed in this article, including compressional wave velocity, effective porosity, hydrocarbon saturation, Young’s modulus and Poisson’s ratio. Using the technique of the operation of artificial neural networks, a prediction of missing information was carried out based on the relationships between the analyzed parameters in the wells, where the estimated data was available. In recent years, there has been a dynamic development of machine learning technology and the so-called artificial intelligence. There are very few fields of science in which they find no application. The hydrocarbon saturation parameter, despite the challenges posed by the interpretation of this parameter, was also subjected to an estimation attempt, confirming the low correlation values between the analyzed parameters and requiring much more advanced work of an individual nature. The results of parametric prediction, previously validated by characterizing the R and RMSE parameters, were applied in the next step in the spatial modeling process of all analyzed parameters. Finally, as part of the visualization of the differences between the use of an incomplete and partially estimated data set in spatial analysis, a map of mean values of the selected parameter within the analyzed interval was presented. The set of data prepared in this way allowed for a more reliable spatial reconstruction of the distribution of parameters important in the context of the characteristics of the hydrocarbon reservoir, on the basis of which, in the subsequent stages, it is possible to more fully assess the deposit potential of the analyzed object. The methodology presented in the article, supported by a real case study, is an alternative to geophysical interpretations that require financial and time resources, sometimes large numbers of boreholes, especially for areas characterized by relatively low spatial variability and tectonic complexity. The condition is the availability of the interpretation in at least several boreholes, constituting a pattern for recreating the variability of the tested parameter / parameters in the remaining profiles of the boreholes.