Neural simulation as a tool for predicting reservoir facies and petrotypes

UDK: 552.124
DOI: 10.24887/0028-2448-2022-1-21-25
Key words: individual functions, lithotypes, neural simulation, neural facies, forecast
Authors: K.A. Belova (Tyumen Petroleum Research Center LLC, RF, Tyumen), N.A. Senachin (Tyumen Petroleum Research Center LLC, RF, Tyumen), O.M. Grechneva (Tyumen Petroleum Research Center LLC, RF, Tyumen), A.A. Snohin (Kynsko-Chaselkoye Neftegas LLC, RF, Tyumen), R.R. Shakirov (Kynsko-Chaselkoye Neftegas LLC, RF, Tyumen), O.A. Loznyuk (Rosneft Oil Company, RF, Moscow)

The paper describes an approach of automation of geological processes analysis by the neural simulation that calls for predicting facies and petrotypes in wells without core and with a limited well logging suite. In the course of the work, non-conventional approaches were applied to use the entire set of input data, including the historical wells stock with partially lost data. The cycle of data analysis and processing, network training and further modeling of facies and lithotypes in wells included the following stages: 1) matching input geological and geophysical data in wells; 2) describing core data and identifying facies that characterize depositional environments, as well as separating lithological types of net-reservoir and non-reservoir; 3) updating petrophysical model taking into account new data. Analyzing and justifying lithological groups and petrotypes identified on well logging data to obtain individual functions of porosity vs. permeability; 4) evaluating the possibility of predicting facies identifiedВ  on core by logging methods based on statistical analysis and neural networks; 5) grouping wells on the basis of existing set of logging curves, building a matrix of training cuttings; 6) training the network, separating facies identified based on the sedimentological analysis of core by logging methods in wells without core; 7) estimating predicted facies and petrotypes in wells by comparing actual and predicted data, as well as by checking with test well that was not included in the training. It is shown, high-quality network training makes it possible to get the right result allowing to use the entire data set, including partially distorted or noisy.


1. Baraboshkin E.Yu., Prakticheskaya sedimentologiya. Terrigennye rezervuary. Posobie po rabote s kernom (Practical sedimentology. Terrigenous reservoirs. On how to operate core samples), Tver': GERS Publ., 2011, 152 p.

2. Zverev K.V, Redina S.A. et al., Seismiofacies and petrofacial modeling of the Sigovskaya formation as a tool for removing uncertainties in the construction of a 3D geological model of the reservoir (In Russ.), PROneft'. Professional'no o nefti, 2019, no. 4(14), pp. 20–25, DOI: 10.24887/2587-7399-2019-4-20-25

3. Sergeev A.P., Tarasov D.A., Vvedenie v neyrosetevoe modelirovanie (Introduction to neural network modeling), Ekaterinburg: Publ. of Ural University, 2017, pp. 26–31.

4. Kachurin S.I., Analiz primenimosti mnogosloynoy neyronnoy seti dlya raspoznavaniya litologicheskoy struktury skvazhiny po dannym geofizicheskikh issledovaniy (Analysis of the applicability of a multilayer neural network for recognizing the lithological structure of a well based on geophysical survey data): thesis of candidate of technical science, Izhevsk, 2003.

5. Rodina S.N., Application of artificial neural networks for well-log data interpretation (In Russ.), Vestnik VGU, 2007, no. 2, pp. 184–188.

To buy the complete text of article (a format - PDF) or to read the material which is in open access only the authorized visitors of the website can. .