Қазақстанның мұнай-газ саласының хабаршысыҚазақстанның мұнай-газ саласының хабаршысы2707-42262957-806XKMG Engineering10901310.54859/kjogi109013Machine Learning-Assisted Prediction of Polymer Flooding Performance in Heterogeneous Carbonate Reservoirs of the Pre-Caspian Basin: A Gradient Boosting and Physics-Informed ApproachZhetpisAdil Almasulyadiljetpeace@gmail.comhttps://orcid.org/0009-0000-0390-5436RyndinVladimir Vitalievichryndin.v@teachers.tou.edu.kzhttps://orcid.org/0000-0002-4248-9516KenzhanovaMensulu Manarbekovnamena-1997@list.ruhttps://orcid.org/0009-0004-4696-6908OmarbekovaInara Kassymkhanovnainara_19@mail.ruhttps://orcid.org/0009-0004-4292-6765AigozhinaDinara Gabdulgazisovnadina7041976@mail.ruhttps://orcid.org/0000-0002-5712-5280AbdullinaGulnara Gosmanovnagulnara_1277@mail.ruhttps://orcid.org/0000-0003-4493-1715830906202625082026Copyright © , Zhetpis A.A., Ryndin V.V., Kenzhanova M.M., Omarbekova I.K., Aigozhina D.G., Abdullina G.G.<p><strong>Background.</strong> Polymer flooding remains one of the most technically viable enhanced oil recovery (EOR) methods for mature fields in Kazakhstan’s Pre-Caspian Basin, yet predicting its sweep efficiency in heterogeneous carbonate reservoirs presents significant challenges due to complex pore-throat geometry, high salinity formation waters, and pronounced vertical permeability contrasts. Conventional reservoir simulation approaches are computationally prohibitive for real-time operational decisions and uncertainty quantification across the required ensemble sizes.</p>
<p><strong>Aim.</strong> This study develops and validates a hybrid machine learning (ML) framework that integrates gradient boosting regression (GBR) with physics-informed constraints derived from polymer transport theory to predict incremental oil recovery factor (ΔRF) and injectivity ratio (IR) for polymer flooding operations in carbonate formations of the Pre-Caspian Basin.</p>
<p><strong>Materials and Methods.</strong> A dataset of 412 polymer flooding pilot and field-scale operations was compiled from published literature, SPE technical reports, and internal field data from three Kazakhstani assets: Tengizchevroil (TCO), Karachaganak Petroleum Operating (KPO), and the Uzen field operated by Ozenmunaigas (OMG). Input features encompass 18 reservoir and fluid parameters. GBR was trained using a stratified 80/20 train-test split with five-fold cross-validation. Physics-informed penalty terms derived from the Darcy-scale polymer transport equations were embedded into the loss function to ensure physical consistency of predictions. Model performance was benchmarked against artificial neural networks (ANN), support vector regression (SVR), and full-physics compositional simulation.</p>
<p><strong>Results.</strong> The GBR model achieved a coefficient of determination and root mean square error RMSE = 2.31% on the held-out test set for ΔRF prediction, outperforming ANN () and SVR (). The physics-informed penalty reduced physically inconsistent predictions (negative injectivity gain) by 78% compared to unconstrained GBR. SHAP (SHapley Additive exPlanations) analysis identified permeability variation coefficient , polymer concentration , and formation water salinity as the three dominant predictors. Incremental recovery factor estimates from the ML surrogate agreed with full compositional simulation within ±1.8% absolute for the Uzen carbonate pilot dataset.</p>
<p><strong>Conclusions.</strong> The proposed physics-informed GBR surrogate model provides rapid and reliable predictions of polymer flooding performance suitable for field-scale workflow integration. The approach substantially reduces the computational burden of EOR screening and optimization for heterogeneous carbonate reservoirs, with direct applicability to ongoing and planned polymer EOR projects in Kazakhstan.</p>polymer floodingmachine learninggradient boostingenhanced oil recoverycarbonate reservoirsPre-Caspian BasinSHAP analysisphysics-informed machine learningKazakhstanpermeability heterogeneityполимерное заводнениемашинное обучениеградиентный бустингповышение нефтеотдачикарбонатные коллекторыПрикаспийский бассейнSHAP-анализфизически-информированное машинное обучениеКазахстаннеоднородность проницаемости[Mirzaev B, Karabassov N, Ergashev T. Petroleum Resources of the Pre-Caspian Basin: Current Status and Development Outlook. Oil Gas J Central Asia. 2023;14(2):18–34. doi:10.32758/ogca.2023.14.2.18][Akhmetov R, Seitkali D, Bekova G. Water flooding performance analysis of mature Kazakhstani carbonate fields. J Pet Sci Eng. 2022;208:109632. doi:10.1016/j.petrol.2021.109632][Tleshev M, Alzhanov N. Enhanced recovery strategies for high-water-cut reservoirs in Mangistau Region. Kazakhstan J Oil Gas Ind. 2024;6(3):22–36. doi:10.54859/kjogi.2024.6.3.22][Lake LW, Johns RT, Rossen WR, Pope GA. Fundamentals of Enhanced Oil Recovery. Tulsa: Society of Petroleum Engineers; 2014.][Sorbie KS. Polymer-Improved Oil Recovery. Glasgow: Blackie Academic & Professional; 1991.][Wang D, Cheng J, Yang Q, Gong W, Li Q. Viscous-elastic polymer can increase micro-scale displacement efficiency in cores. SPE J. 2000;5(3):281–292. doi:10.2118/63227-PA][Sheng J. Enhanced Oil Recovery Field Case Studies. Amsterdam: Elsevier; 2013.][Christie MA, Blunt MJ. Tenth SPE comparative solution project: A comparison of upscaling techniques. SPE Reserv Eval Eng. 2001;4(4):308–317. doi:10.2118/72469-PA][Al-Dousari MM, Garrouch AA. An artificial neural network model for predicting the recovery performance of surfactant polymer floods. J Pet Sci Eng. 2013;109:187–198. doi:10.1016/j.petrol.2013.08.024][Mogensen K, Masalmeh S. A review of EOR techniques for carbonate reservoirs in challenging reservoir conditions. J Pet Sci Eng. 2020;195:107889. doi:10.1016/j.petrol.2020.107889][Chen T, Guestrin C. XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016:785–794. doi:10.1145/2939672.2939785][Lundberg SM, Lee SI. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst. 2017;30:4765–4774.][Zhu D, Bai B, Hou J. Polymer gel systems for water management in high-temperature and high-salinity oil reservoirs. Ind Eng Chem Res. 2017;56(29):7904–7916. doi:10.1021/acs.iecr.7b01562][Raissi M, Perdikaris P, Karniadakis GE. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys. 2019;378:686–707. doi:10.1016/j.jcp.2018.10.045][Karniadakis GE, Kevrekidis IG, Lu L, et al. Physics-informed machine learning. Nat Rev Phys. 2021;3(6):422–440. doi:10.1038/s42254-021-00314-5][Dake LP. Fundamentals of Reservoir Engineering. Amsterdam: Elsevier; 1978.][Dykstra H, Parsons RL. The prediction of oil recovery by waterflood. In: Secondary Recovery of Oil in the United States. 2nd ed. Washington, DC: API; 1950:160–174.][Stiles WE. Use of permeability distribution in water flood calculations. Trans AIME. 1949;186:9–13.][Goudarzi A, Delshad M, Sepehrnoori K. A chemical EOR benchmark study of different reservoir simulators. Comput Geosci. 2016;20(2):469–482. doi:10.1007/s10596-015-9526-z][Seitenova GZ, Nurmagambetov AS, Iskakova ZB. HPAM adsorption and viscosity retention in Tengiz-type dolomite cores under high-salinity conditions. Kazakhstan J Oil Gas Ind. 2023;5(4):45–58. doi:10.54859/kjogi.2023.5.4.45]