«Oilfield engineering»

Rock properties petrophysical prediction using machine learning algorithms

UDC: 550.8.053+550.8.056
DOI: -

Authors:

¹ Ufa University of Science and Technology, Ufa, Russia

Keywords: machine learning, petrophysical properties prediction, acoustic logging well, logging interpretation, random forest method, deep neural networks

Annotation:

Currently, machine learning is an active powerful instrument that has been integrated into the oil and gas industry and has become a part of it in case of working with the results of geophysical well surveys. Machine learning methods can speed up (reduce time and eliminate human factor) the process of working with well data both at the stage of preliminary processing and normalization of "raw" curves and at the stage of interpretation (for example, determining working intervals based on the results of geophysical well surveys or lithophysical types based on well logs). In addition, methods are being studied to improve the results of geophysical well surveys at the qualitative level (smoothing of "noisy" curves by various filtering methods, their recovery in the no-data intervals, etc.). All these facts make machine learning one of the prior and most developed directions in the oil and gas industry. Nowadays, to solve different problems (reservoirs spreading prediction, wellbore stability, etc.) it is required to use machine learning models for petrophysical properties restoration of productive intervals (such as rock density and interval time).

The object of the study is a group of vertical wells of the Orenburg field, where a set of geophysical investigations was performed using various methods: radioactive (gamma and gamma-gamma ray), electrical (lateral and micro-lateral logging, induction logging, etc.), neutron (neutron gamma logging, neutron-neutron logging) and acoustic (acoustic logging). The initial data arrays contain the results of downhole measurements by depth.

The purpose of this work is to build a mechanism for predicting (forecasting) petrophysical properties of rocks using machine learning methods on the example of a borehole with incomplete well logging data.

The result of this study is a prediction method (mechanism) that allows automated analyzing arrays with recorded logging data, detecting depth intervals with missing data in these arrays and predicting the shape of well logging data using Random Forest machine learning algorithm.

Bibliography:

1. Enhanced Reservoir Characterization for Optimizing Completion Decisions in the Permian Basin Using a Novel Field-Scale Workflow including Wells with Missing Data / A.P. Garcia, L.M. Hernandez, A. Jagadisan [et al.] // Unconventional Resources Technology Conference, Denver (Colorado, USA), 22–24 July 2019. – DOI: 10.15530/urtec-2019-317

2. Primenenie metodov mashinnogo obucheniya v obrabotke dannykh geofizicheskikh issledovaniy skvazhin otlozheniy vikulovskoy svity / V.I. Sakhnyuk, E.V. Novikov, A.M. Sharifullin [i dr.] // Georesursy. – 2022. – № 24(2). – S. 230–238. – DOI: 10.18599/grs.2022.2.21

3. Cuddy S. Using Fuzziness to Repair Borehole Electrical Logs // InSite. CWLS Magazine. – 2006. – № 5. – Р. 9–12.

4. Vakhitova G.R., Kasimov A.M. Avtomaticheskoe vosstanovlenie karotazhnykh diagramm razlichnymi metodami approksimatsii // Geonauki: vremya peremen, vremya perspektiv: sb. st. v GeoLib 10-y Mezhdunar. konf. "Geomodel’". – SPb., 2023. – S. 467–470.

5. Accelerating and Enhancing Petrophysical Analysis with Machine Learning: A Case Study of an Automated System for Well Log Outlier Detection and Reconstruction / R. Akkurt, T.T. Conroy, D. Psaila [et al.] // SPWLA 59th Annual Logging Symposium, London, UK, 2–6 June 2018. – SPWLA-2018-BB.

6. Shier D. Well log normalization: Methods and guidelines // The SPWLA J. of Formation Evaluation and Reservoir Description. – 2004. – Vol. 45(3). – P. 268–280.

7. Merembayev T., Yunussov R., Amirgaliyev Y. Machine Learning Algorithms for Classification Geology Data from Well Logging // 14th International Conference on Electronics Computer and Computation (ICECCO), Kaskelen, Kazakhstan, 29 Nov. – 1 Dec. 2018. – P. 206–212. – DOI: 10.1109/ICECCO.2018.8634775

8. Formation Lithology Classification: Insights into Machine Learning Methods (SPE-196096-MS) / I.M. Mohamed, S.M. Mohamed, I. Mazher, Ch. Pieprzica // SPE Annual Technical Conference and Exhibition, Calgary, Alberta, Canada, 23 Sept. 2019. – DOI: 10.2118/196096-MS

9. Hydrocarbon Index Identification in Lateral Section of Horizontal Wells Using Machine Learning (SPE-201382-MS) / M. Toktarov, M.A. Siddiqui, I. Permata, S. Khakimov // SPE Annual Technical Conference & Exhibition, Denver, Colorado, USA, 5–7 Oct. 2020. – DOI: 10.2118/201382-MS

10. Automatic Interpretation of Well Logs with Lithology-Specific Deep-Learning Methods / A.P. Peyret, J.A. Garrido, C. Torres-Verdin, J. Strobel // SPWLA 60th Logging Symposium, the Woodlands, Texas, USA, 15–19 June 2019. – DOI: 10.30632/T60ALS-2019_SSSS

11. Machine learning-based method for automated well log processing and interpretation / P. Wu, V. Jain, V. Kulkarni, A. Abubakar // SEG International Exposition and Annual Meeting, Anaheim, California, USA, 14 Oct. 2018. – Р. 2041–2045. – DOI: 10.1190/segam2018-2996973.1