Artificial Neural Network-Based Predictive Model Development for Reservoir Rock Permeability
DOI:
https://doi.org/10.3329/cerb.v23i10.78502Keywords:
Data analytics, Machine learning, Model accuracy, Rock permeability, Reservoir characterizationAbstract
Rock permeability is an important parameter for reservoir quality assessment of any hydrocarbon reservoir. The traditional methods for determining permeability include core analysis and well-test techniques. These approaches are time-consuming and costly. As a result, various studies have been conducted to predict rock permeability using core and log data with machine learning approaches. The aims of the study are to investigate the performance of data-driven predictive models in determining rock permeability and analysis of model accuracy. In the study, 260 log data sets from a gas field in the Bengal basin are adopted to forecast the reservoir rock permeability using an artificial neural network (ANN). Using the most suitable parameters, the data set was divided into three distinct categories such as 60% for training, 20% for testing, and 20% for validation. The most common two algorithms of Levenberg-Marquardt (LM), and Bayesian regularization (BR) have been applied in determining permeability to train the ANN-based model. The LM algorithm training procedure delivers the best match between the target and predicted values of permeability using the predictor variables (such as sonic travel time, gamma ray, bulk density, formation resistivity, and neutron porosity, compared to the BR-based optimized ANN network strategies. For instance, the LM algorithm provides an excellent outcome, which has a correlation coefficient (%) and average absolute percentage error of 89.90 and 6.54, compared to the BR algorithm of 67.55 and 10.93 for testing data sets, respectively. The studied procedures of the ANN-based model can be applied to predict the penetration rate of drilling, reservoir rock quality assessment and oil recovery prediction for reservoir simulation studies.
Chemical Engineering Research Bulletin 23(2023): 36-40
Downloads
56
47