Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yerzhan Kenzhebek has a total of 26 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Separating Artificial-Viscosity Annealing from the L-BFGS Viscosity in Physics-Informed Neural Networks for Two-Phase Flow

    by Saltanbek Mukhambetzhanov, Yerzhan Kenzhebek, Samson Dawit Bekele, Saida Tastanova, Dagmawi Lemma, Timur Imankulov

    Physics-informed neural networks (PINNs) remain difficult to train for hyperbolic conservation laws. Artificial viscosity is a common remedy, and several methods reduce its coefficient during training. We examine adaptive moment estimation (Adam) followed by limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) optimisation. The schedule followed during Adam, the viscosity used during L-BFGS, and the checkpoint supplied between them can change together. We separate these factors for the one-dimensional Buckley–Leverett equation using a fixed multilayer perceptron with hyperbolic-tangent activations and 15 paired seeds per configuration. The central experiments impose a zero-diffusive-flux outlet condition. Lowering the L-BFGS viscosity from 3.5×10−3 to 1.67×10−4 increased full-grid relative L2 error by 0.0413±0.0069. The schedule–viscosity interaction was −0.0067±0.0075 and was not detected after multiplicity adjustment. A decay ending at 3.5×10−3 improved on the tested low-endpoint decay before high-viscosity polishing, while its difference from constant high viscosity remained inconclusive. The higher of the two tested constant viscosities improved aggregate and pre-breakthrough accuracy but increased post-breakthrough error. Stage-matched curricula showed no detected accuracy advantage; their greater executed cost arose from intermediate diagnostic L-BFGS phases whose states were discarded. These results identify the terminal viscosity, checkpoint rule, outlet treatment, and optimiser exposure as controls needed to interpret schedule comparisons in this pipeline.

  • Open Access

    Identifiability and Conditioning of Flux-Parameter Inversion for the Buckley–Leverett Equation

    by Samson Dawit Bekele, Yerzhan Kenzhebek, Saltanbek Mukhambetzhanov, Saida Tastanova, Dagmawi Lemma, Timur Imankulov

    We analyse inversion of the viscosity ratio M in the Buckley–Leverett equation, distinguishing structural identifiability, observation-operator conditioning, and numerical estimation error. The Riemann solution gives M=4s2−4s from the shock speed s, with relative condition number κ(M)=21+M/(1+M−1), which diverges as M↓0. The continuum physics objective admits exact solution–parameter pairs for every M>0, so identifying the true ratio requires observations. We examine numerical recovery using a viscous physics-informed neural network (PINN) with positive parameterisation and 30 paired repetitions per setting. At M=3, the median parameter error decreases from 514.25% without observations to 6.13% with 25 noisy observations. Additional data decrease the median field error, while median parameter errors are nonmonotone. Over M∈[0.25,10], κ decreases from 18.94 to 2.86, whereas the smooth-branch Cramér–Rao relative scale for 50 scattered observations with 5% Gaussian noise levels off near 7%. Shock crossings supply additional nonregular information. Finite-network physics losses remain small across the tested ratios, with different orderings on training and held-out points. Scalar controls assess finite-resolution viscous-model discrepancy; complementary PINN tests show poorer recovery as the front sharpens. Recovery accuracy therefore depends jointly on the observation operator, regularisation, and numerical estimator.

  • Open Access

    Physics-Informed Neural Networks for Coupled Non-Isothermal Two-Phase Flow in Porous Media: Training Pathologies, Remedies, and the Role of Gradient Conflict

    by Yerzhan Kenzhebek, Timur Imankulov, Samson Dawit Bekele, Saltanbek Mukhambetzhanov

    Physics-informed neural networks (PINNs) are difficult to train on strongly coupled, multi-objective systems: with fixed loss weights an otherwise identical run succeeds or diverges with the random seed, and a leading explanation attributes the failures to conflict between the loss-term gradients. We test that explanation on a coupled non-isothermal Buckley–Leverett displacement with a passive tracer, a three-front hot-water flood with an exact analytical solution as ground truth. The test produces an exactly verified counterexample: the conflict-free update ConFIG attains its alignment objective yet converges to a stable solution that violates the equations, while a quasi-second-order optimizer (SOAP) and per-term gradient surgery (PCGrad) train the network reliably with no loss-weight search. The most aligned update is thus the one that fails: gradient alignment is neither necessary nor sufficient for trainability, and the conflict is a diagnostic of the loss rather than its cause. The same formulation serves the forward, parametric, and inverse problems: it predicts the movable-oil recovery with a relative error of about 5%, returns the solution for any oil viscosity in a trained range from one network, and, from sparse near-injector monitoring, recovers the thermal front speed and one transport coefficient when the other is known. To our knowledge this is the first PINN solution of this coupled displacement, which is central to thermal oil recovery.


  • Open Access

    A Systematic Analysis of Physics-Informed Neural Networks for Two-Phase Flow with Capillarity: The Muskat–Leverett Problem

    by Timur Imankulov, Alibek Kuljabekov, Samson Dawit Bekele, Zhumabek Zhantayev, Bakytzhan Assilbekov, Yerzhan Kenzhebek

    This work develops and systematically evaluates a physics-informed neural network (PINN) solver for the fully coupled, time-dependent Muskat–Leverett system with capillarity modeled in the pressure equation. A single shallow–wide multilayer perceptron jointly predicts wetting pressure and water saturation; physical capillary pressure regularizes the saturation front, while a small numerical diffusion term in the saturation residual acts as a training stabilizer rather than a shock-capturing device. To guarantee admissible states in stiff regimes, we introduce a saturation soft-clamping head enforcing 0<Sw<1 and activate it selectively for stiff mobility ratios. Using IMPES solutions as reference, we perform a sensitivity study over network depth and width, interior collocation and boundary data density, mobility ratio, and injection pressure. Shallow-wide networks (10 layers × 50 neurons) consistently outperform deeper architectures, and increasing interior collocation points from 5000 to 50,000 reduces mean saturation error by about half, whereas additional boundary data have a much weaker effect. Accuracy is highest at an intermediate mobility ratio and improves monotonically with higher injection pressure, which sharpens yet better conditions the front. Across all regimes, pressure trains easily while saturation determines model selection, and the PINN serves as a physics-consistent surrogate for what-if studies in two-phase porous-media flow.


  • Efficiency of Parallel Machine Learning Algorithms on Multiprocessor and GPU Architectures

    by Maksat Mustafin, Madina Mansurova, Aida Zhakyp, Aksultan Mukhanbet, Yerzhan Kenzhebek, Nurislam Kassymbek

  • Open Access

    Enhancing Oil Recovery Predictions by Leveraging Polymer Flooding Simulations and Machine Learning Models on a Large-Scale Synthetic Dataset

    by Timur Imankulov, Yerzhan Kenzhebek, Samson Dawit Bekele, Erlan Makhmut

    Polymer flooding is a prominent enhanced oil recovery process that is widely recognized for its cost-effectiveness and substantial success in increasing oil production. In this study, the Buckley–Leverett mathematical model for polymer flooding was used to generate more than 163,000 samples that reflect different reservoir conditions using seven input parameters. We introduced artificial noise into the dataset to simulate real-world conditions and mitigate overfitting. Seven classic machine learning models and two neural networks were trained on this dataset to predict the oil recovery factor based on the input parameters. Among these, polynomial regression performed best with a coefficient of determination (R2) of 0.909, and the dense neural network and cascade-forward neural network achieved R2 scores of 0.908 and 0.906, respectively. Our analysis included permutation feature importance and metrics analysis, where key features across all models were identified, and the model’s performance was evaluated on a range of metrics. Compared with similar studies, this research uses a significantly larger and more realistic synthetic dataset that explores a broader spectrum of machine learning models. Thus, when applied to a real dataset, our methodology can aid in decision-making by identifying key parameters that enhance oil production and predicting the oil recovery factor given specific parameter values.


  • Open Access

    Implementation of regression algorithms for oil recovery prediction

    by Yerzhan Kenzhebek, Timur Imankulov, Darkhan Akhmed-Zaki, Beimbet Daribayev

    This paper presents the work of predicting oil production using machine learning methods. As a machine learning method, a multiple linear regression algorithm with polynomial properties was implemented. Regression algorithms are suitable and workable methods for predicting oil production based on a data-driven approach. The synthetic dataset was obtained using the Buckley-Leverett mathematical model, which is used to calculate hydrodynamics and determine the saturation distribution in oil production problems. Various combinations of parameters of the oil production problem were chosen, where porosity, viscosity of the oil phase and absolute permeability of the rock were taken as input parameters for machine learning. And the value of the oil recovery factor was chosen as the output parameter. More than 400 thousand synthetic data were used to test multiple regression algorithms. To estimate the quality of regression algorithms, the mean square error metrics and the coefficient of determination were used. It was found that linear regression does not cover all patterns in the data due to underfitting. Various degrees of polynomial regression were deployed and tested, and it was also found that for our synthetic data, the quadratic polynomial model trains quite well and perfectly predicts the value of the oil recovery factor. To solve the overfitting problem, L1 regularization known as the Lasso regression method was applied. For the quadratic polynomial regression model, the coefficient of determination was 0.96, which is a pretty good result for the test data. Thus, it is assumed that the data-driven machine learning methods discussed in the paper can be useful for predicting the oil recovery factor using practical data from oil fields at the stages of production


  • Using Machine Learning Algorithms to Solve Polymer Flooding Problem

    by T. Imankulov, Y. Kenzhebek, E. Makhmut, D. Akhmed-Zaki


  • Using Machine Learning Methods for Oil Recovery Prediction

    by B. Daribayev, D. Akhmed-Zaki, T. Imankulov, Y. Nurakhov, Y. Kenzhebek