Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Uchenna Odi has a total of 33 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Automated Production Segmentation of Rate-Time Data for Improved Decline-Curve Forecasting

    by Uchenna Odi, Karri Reddy, Cenk Temizel, Muhammad Alhotan

    Production forecasting is a critical step in developing type curves and estimating reserves, but it becomes unreliable when a well’s rate-time history departs from a single decline trend — because of choke management, parent-child interference, and other operational impacts — since fitting one decline curve across such a history biases the resulting estimated ultimate recovery (EUR). This paper presents a data-driven segmentation workflow that identifies decline-regime boundaries directly from rate-time data, without operator-supplied event flags, using a rolling-window outlier test on windowed rate and the Bourdet-derived loss ratio and b-value. Each segment is fit with six empirical decline models, and the fitted parameters are clustered to label operational regimes; a parallel benchmark compares these fits against XGBoost and LSTM forecasts. On a 60-well demonstration population across the Eagle Ford, the algorithm resolved 1 to 12 segments per well (mean 4.6), and cross-model agreement in gas EUR improved with segmentation: the coefficient of variation across the six empirical models fell from 0.28 for single-segment wells to 0.07 for wells resolved to six or more segments (r=-0.49, n=60 ). Under-resolved wells were also where the machine-learning benchmark was least stable, an effect traced to two wells with missing information. Extending the analysis to a 300-well population, (i) gas EUR rose roughly nine-fold from the most liquids-rich to the driest sub-play, consistent with known geology; (ii) the segmentation workflow, used unmodified as a parent-child interference detector, distinguished interference-attributable segment boundaries from other-cause boundaries with 89% held-out accuracy; and (iii) the full eight-method benchmark, including LSTM, reproduced the same tightening with markedly improved stability relative to the smaller population. Cross-phase agreement between gas and oil was positive but incomplete (r=0.38), traced largely to phase-specific signal strength. The workflow requires no manual picks or event catalog and can be applied unattended at field scale.

  • Open Access

    Forecasting Hydrocarbon Production in Unconventional Shale Using Deep Learning and Gradient Boosting

    by Veena Kumar, Siddharth Misra, Uchenna Odi, Andrew Silver, Ola Alhulaimi, Ebtehal Alharbi, Meshal Alaithan, Hesham Alshahrani, Clarissa Cemer, Jameel Muallim, et al.

    Accurate long-term production forecasting is critical for optimizing recovery strategies and economic evaluations in unconventional reservoirs, where traditional decline curve analysis often struggles with the complexities of modern completions and inherent reservoir heterogeneity. This study addresses these challenges by comparing the performance of three deep learning workflows and one traditional machine learning approach in forecasting five years of monthly gas and condensate production from horizontal shale wells. The deep learning models include Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks (both with and without Non-negative Matrix Factorization (NMF) for feature reduction), while the traditional machine learning method is Extreme Gradient Boosting (XGBoost).,Each of the four forecasting workflows was developed and tuned independently by separate teams using an identical dataset, ensuring a consistent and unbiased basis for evaluation. Model performances were rigorously assessed based on prediction accuracy, overfitting tendencies, and interpretability. Our findings reveal that deep learning models, particularly those incorporating advanced feature engineering techniques like dimensionality reduction, offer significant advantages in capturing complex production trends. These models demonstrated a remarkable ability to minimize both bias and variance, leading to more robust and reliable forecasts compared to traditional methods. Specifically, the integration of NMF with LSTM networks showed considerable promise in handling the high-dimensional and often correlated features inherent in production data obtaining the best accuracy, with Mean Absolute Percentage Error (MAPE) of 3.04% (gas) and 3.96% (condensate).,This comprehensive comparative study provides a practical and valuable benchmark for selecting data-driven forecasting approaches in the development of unconventional assets. It offers crucial insights into the trade-offs between different deep learning architectures and boosting methods in real-world deployment scenarios, empowering practitioners to make informed decisions for enhancing production forecasting accuracy and optimizing economic outcomes in complex unconventional plays.

  • Open Access

    Permeability Prediction Using Vision Transformers

    by Cenk Temizel, Uchenna Odi, Kehao Li, Lei Liu, Salih Tutun, Javier Santos

    Accurate permeability predictions remain pivotal for understanding fluid flow in porous media, influencing crucial operations across petroleum engineering, hydrogeology, and related fields. Traditional approaches, while robust, often grapple with the inherent heterogeneity of reservoir rocks. With the advent of deep learning, convolutional neural networks (CNNs) have emerged as potent tools in image-based permeability estimation, capitalizing on micro-CT scans and digital rock imagery. This paper introduces a novel paradigm, employing vision transformers (ViTs)—a recent advancement in computer vision—for this crucial task. ViTs, which segment images into fixed-sized patches and process them through transformer architectures, present a promising alternative to CNNs. We present a methodology for implementing ViTs for permeability prediction, its results on diverse rock samples, and a comparison against conventional CNNs. The prediction results suggest that, with adequate training data, ViTs can match or surpass the predictive accuracy of CNNs, especially in rocks exhibiting significant heterogeneity. This study underscores the potential of ViTs as an innovative tool in permeability prediction, paving the way for further research and integration into mainstream reservoir characterization workflows.


  • Open Access

    Classifying Facies in 3D Digital Rock Images Using Supervised and Unsupervised Approaches

    by Cenk Temizel, Uchenna Odi, Karthik Balaji, Hakki Aydin, Javier E. Santos

    Lithology is one of the critical parameters influencing drilling operations and reservoir production behavior. Well completion is another important area where facies type has a crucial influence on fracture propagation. Geological formations are highly heterogeneous systems that require extensive evaluation with sophisticated approaches. Classification of facies is a critical approach to characterizing different depositional systems. Image classification is implemented as a quick and easy method to detect different facies groups. Artificial intelligence (AI) algorithms are efficiently used to categorize geological formations in a large dataset. This study involves the classification of different facies with various supervised and unsupervised learning algorithms. The dataset for training and testing was retrieved from a digital rock database published in the data brief. The study showed that supervised algorithms provided more accurate results than unsupervised algorithms. In this study, the extreme gradient boosted tree regressor was found to be the best algorithm for facies classification for the synthetic digital rocks.


  • Experimental Investigation of Wet Gas Dew Point Pressure Change with Carbon Dioxide Concentration

    by Uchenna Odi, Hicham El Hajj, Anuj Gupta

    Dew point pressure is a critical measurement for any wet gas reservoir. Condensate blockage is likely when the reservoir pressure decreases below the dew point pressure and this can result in a reduction of gas productivity. Errors in measuring dew point pressure can lead to errors in the estimation of the onset of condensate blockage and thus be detrimental to the management of wet gas fields. This work presents experimental verification of a new method of determining dew point pressures for wet gas fluids. Results obtained from this method are compared to calculated values based on Peng Robinson equation of state. Dew point pressure determination is important when devising solutions on how to prevent condensate blockage. One possible treatment fluid, carbon dioxide, has the ability to lower dew point pressures and thus delay the onset of condensate blockage. The novel method presented in this work was applied to determine the experimental dew point pressure of several wet gas mixtures as a function of carbon dioxide concentration. These experiments also show the potential of using carbon dioxide to lower dew point pressures in wet gas fields. Experimental results show close match between the experimental estimates of dew point pressure and the Peng Robinson calculations. Experimental results also support the general observation that carbon dioxide has the ability to lower the dew point pressure of wet gas fields. The results of this work are useful in Enhanced Oil/Gas Recovery processes that utilize carbon dioxide and for Huff and Puff which uses carbon dioxide to remove and prevent further build-up of condensate banks in wet gas reservoirs. This work investigates experimental conditions showing the change in dew point pressure as a function of carbon dioxide concentration. This dynamic relationship can be used to tune equation of state models which, in turn, allows more accurate reservoir modeling of hydrocarbon recovery process.


  • Visualization and Analysis of Interaction of Carbonate Reservoirs with Carbon Dioxide During CO2 Enhanced Gas Recovery Process

    by Uchenna Odi, Anuj Gupta

    This paper presents simulation based analysis for CO2 Huff-n-puff and Enhanced Gas Recovery (EGR) process in carbonate rocks that are representative of the Middle East’s aquifers and oil/gas reservoirs. The objective of this study is to study the development of condensate in the near wellbore region of a wet gas carbonate reservoir and to subsequently study the feasibility of using CO2 injection for removing this condensate phase. Changes in oil saturation, gas saturation, gas permeability, and gas productivity were used to quantify the effect that CO2 has on wet gas carbonate reservoirs. Carbonate reservoir systems have complex structures and to represente these, two reservoir models were constructed that had the characteristics of typical carbonate reservoirs. Condensates appeared in wet gas systems once the pressure reduces below the dew point pressure. This can be remedied in two ways: (1) repressurizing the reservoir above the dew point and (2) lowering the dewpoint pressure by changing the fluid properties of the reservoir fluid. Both of these remedies can be achieved by using CO2. Results show that CO2 has the ability to repressurize a reservoir and the ability to lower the dew point of a wet gas fluid. Results also show that carefully timing the start of CO2 injection and its duration can significantly improve the economics of enhanced gas recovery process.


  • Optimization and Design of Carbon Dioxide Flooding

    by Uchenna Odi, Anuj Gupta

    Increasing energy demand coupled with public concern for the environment has placed the oil industry in an awkward position as profit-making energy provider and protector of the environmental. Today fossil fuels are the most efficient at meeting the global energy needs but have the undesirable outcome of potentially harming the environment due to production of carbon dioxide. Carbon dioxide flooding of oil reservoirs is a proven method for increasing oil production by using the miscible interactions between reservoir oil and carbon dioxide. Carbon dioxide flooding is beneficial to the environment and to petroleum producers because it uses the harmful carbon dioxide as the main agent to increase oil production. The main challenge in using this technology is applying laboratory data to its design and optimization. The objective of this work is to develop a strategy for applying laboratory data to the design and optimization of carbon dioxide flooding process. To accomplish this, reservoir simulation studies were conducted based on carbon dioxide laboratory data. Carbon dioxide laboratory core flood results were studied and modeled using a reservoir simulator. Once there was a match between laboratory data and core flood simulations the results were used to simulate field scaled carbon dioxide flooding. Results of these simulation studies were used to perform optimization using adjoint based algorithms and non adjoint based algorithms. Adjoint based optimization algorithms require access to simulator code while non adjoint based algorithms do not require access to simulator code. A comparison of both methods is predented in this work to show their advantages and disadvantages. This work will address best practices in applying laboratory data to simulation and optimization studies. The work-flow presented in this work can be used to design and optimize field carbon dioxide floods. These results will serve as a basic guideline in applying laboratory data to full field carbon dioxide flood optimization.