Currently number of oil and gas field with easy recoverable reserves is decreasing dramatically. Therefore oil companies consider assets with complex geological conditions and wide range of uncertainty more frequently. Eastern Siberia carbonate deposits can be considered as such complex fields where prediction of geological and petrophysical properties can be a very challenging task due to sedimentary processes, tectonics and following secondary processes. Conventional approaches and methods for reservoir properties forecasting and field development strategy optimization almost cannot be applied for those fields due to they are not capable to take into account all complex geological factors. In such a situation development of new methods, approaches and tools for properties estimation capable to integrate whole amount of available geological data become a crucial necessity. Machine learning algorithms can act as such a tool due to their ability to handle any amount of information and capability for solving different oil and gas task was already confirmed by many researches in last few years.
The main goal of presented work was testing of machine learning algorithms applicability in tasks of wells productivity index forecasting for complex oilfields. Wide range of different factors was tested as features for proposed machine learning models. Those factors included geological, tectonic, stratigraphic and technological (well drilling and completion) data. Presented method could help reservoir engineer to optimize field development strategy not only via improvement of well productivity forecasting robustness but also by exploration of new complex relationships within data which could be not obvious for specialist.
1. Shiwei Yu, Kejun Zhu, Fengqin Diao, A dynamic all parameters adaptive BP neural networks model and its application on oil reservoir prediction, Applied Mathematics and Computation, 2008, V. 195, pp. 66–75.
2. Belozerov B., Bukhanov N., Egorov D., Zakirov A. et al., Automatic well log analysis across Priobskoe field using machine learning methods (In Russ.), SPE 191604-18RPTC-MS, 2018.
3. Cawley G.C., Talbot N.L.C., On over-fitting in model selection and subsequent selection bias in performance evaluation, JMLR, 2010, no. 11(Jul), pp. 2079−2107.
4. Zhihua Zhou, On the doubt about margin explanation of boosting, Artificial Intelligence, 2013, V. 203, pp. 1–18.
5. Chawla V.N., Bowye W.K., Hall O.L., Kegelmeyer W.Ph., SMOTE: Synthetic minority over-sampling technique, Journal of Artiﬁcial Intelligence Research, 2002, V. 16, pp. 321–335.
6. David M.W., Powers evaluation: From precision, recall and F-measure to ROC, informedness, markedness & correlation, Journal of Machine Learning Technologies, 2011, V. 2 (1), pp. 37–63.