Research
In this work, we aim to understand the influence of various reservoir and well parameters on the amount of hydrogen produced during the in-situ hydrogen production process. To do this, we applied the framework developed in our previous work (Ikpeka & Ugwu, 2023), on a simple reservoir model. We developed a proxy model using Box Benkhen Design of Experiment, and implemented a genetic algorithm based optimation on the proxy model.
In this study, we developed a framework to design an in situ combustion model, taking into account four key hydrogen-forming reactions: steam reforming, partial oxidation, autothermal reforming, and pyrolysis. Using Newton-Jacobi iteration, I solved a set of non-linear equations derived from the chemical equilibrium analysis of these reactions. By analyzing the change in Gibbs free energy for each reaction, I was able to screen and implement a numerical model. I then validated the results from this combustion model against those from the thermal reservoir simulator CMG STARS. The model showed a sinusoidal upward trend between the steam-carbon ratio and the amount of hydrogen produced. This combustion model can serve as a framework for designing experimental analyse
Liquid loading of gas wells occurs when liquid condensates accumulate in the wellbore during production. It causes several production challenges and could ultimately kill the well.
In this work, Michael and I developed a new and improved model for predicting liquid loading in gas wells and integrated it into Microsoft Excel. I introduced a deformation coefficient "C" to account for the changes in the shape of liquid droplets as they move along the wellbore. This helps predict the critical rate more accurately when the droplets change from spherical to flat shapes. Our analysis showed that Turner's model had a 35% error, Li's model had a 26% error, and our new model only had a 20% error. Overall, our new model proved to be more accurate in predicting the onset of liquid loading.
In this study we analyzed the properties of reservoir rock and fluid, along with selected well parameters, to create decision-based models that predict initial gas production rates for tight gas formations. Using two machine learning models (ANN and GLM) and production data from 224 wells, we were able to predict the expected recovery rate of new wells. Then, we compared the model predictions with the actual initial gas production rates. The ANN model had a Mean Square Error (MSE) of 1.24, while the GLM model had an MSE of 1.57. The key performance indicators showed that reservoir thickness was the most significant factor for the ANN model, contributing 36.5% to the initial gas production rate, followed by the flowback rate at 29%. Overall, reservoir and fluid properties contributed 53% to the initial gas production rate, while hydraulic fracture parameters contributed 47%.
In this study, I looked at how well machine learning models can predict the dew point pressure in gas condensate reservoirs. I used 535 experimental data points, with temperatures up to 304°F and pressures up to 10,500 psi. First, I used a standard multiple linear regression (MLR) as a baseline to compare the machine learning models. I tried different models: Multilayer Perceptron Neural Networks (MLP), Support Vector Machine (SVM) with a radial basis function kernel, and Decision Trees using Gradient Boost Method (GBM) and XGBoost (XGB). I compared their performance to other published models. For my inputs, I included gas composition, specific gravity, molecular weight of the heavier component, and compressibility factor. I evaluated how well these algorithms worked using root mean square error (RMSE), absolute average relative deviation percentage (AARD%), and the coefficient of determination (R2).
Previous studies show that direct current can improve fluid flow in pore spaces, reduce water production, and decrease hydrogen sulfide without harming the environment. This process is known as Electrokinetic Enhanced Oil Recovery (EK-EOR). However, while past studies show mixed success of the EK-EOR technique, its exact mechanisms and effectiveness are still uncertain.
To get a clearer picture, I conducted a systematic literature review, analyzing 52 articles following the PRISMA protocol. I used data from these articles to run a Monte Carlo simulation with 10,000 iterations to gauge the success rate of EK-EOR, which turned out to be only 45%. I found that the presence of interstitial clay in reservoir rocks impacts the electro-osmotic permeability, which is crucial for EK-EOR's effectiveness. Additionally, salt build-up on the cathode and gas generation at the anode (oxygen and chlorine) are major drawbacks of this technique. My article highlights these issues and suggests areas for future research and application of EK-EOR .
