Porosity and Permeability Prediction of Oil and Gas Reservoirs Using Artificial Neural Networks and Support Vector Machines

This paper focuses on the prediction of porosity and permeability of oil and gas formations using support vector machine and artificial neural networks. The goal is to improve the accuracy and reliability of oil and gas formation porosity and permeability predictions by comparing the performance of these models with traditional statistical methods using oil and gas petrophysical well data. Also in the study, processing methods for extracting data features of well log data consisting of various petrophysical and geological parameters are considered. Research results are presented through detailed graphical visualizations, error distributions, scatter plots, and performance metrics. The performance and accuracy of each model were evaluated using statistical measures such as mean squared error (MSE), R2, and mean absolute error (MAE). The artificial neural network model gave better accuracy than support vector machine and traditional regression methods. The reduction in prediction errors demonstrated by these models provides greater confidence in operational decision-making. Accurate predictions from these models help increase confidence in oil and gas decision-making and reduce operational costs and resource wastage.

I. Introduction

The main petrophysical properties of oil and gas formations are porosity and permeability, which are crucial for hydrocarbon storage and processing. These parameters determine fluid storage and transmission of the formation, which is an important factor in controlling fluid flow, optimizing production, and effectively locating the well. Therefore, accurate and reliable prediction of these characteristics is an important task that ensures the rapid development of the oil and gas industry. Traditional approaches such as core sample analysis, empirical correlations, and the use of well log data are laborious and time-consuming processes that do not provide sufficient efficiency to account for the complex heterogeneous structures of reserves [1]. These methods are often insufficient for accurate predictions due to the variability of the data, which leads to difficulties in planning recovery strategies and efficient well placement. Therefore, artificial intelligence technologies, machine learning algorithms and artificial neural networks offer a modern and efficient approach to predict the petrophysical properties of oil and gas fields [2]. The purpose of this study is to develop and evaluate the effectiveness of new methods of Support Vector Machines (SVM) and Artificial Neural Networks (ANN) models in predicting the porosity and permeability of oil and gas reservoirs based on well log data. It provides extensive validation of the applicability of these models, as well as the development of high-performance predictive models using artificial intelligence tools and their industrial application [3]. The study outlines broad analytical directions for the application of artificial intelligence models in improving the reservoir characterization process in the oil and gas industry. In the past few years, several studies have investigated different machine learning algorithms for predicting reservoir properties such as porosity and permeability [4]. Tahmasebi and Hezarkhani (2019) investigated the application of machine learning algorithms such as support vector machines and neural networks to predict petrophysical properties from well log data and showed improved accuracy compared to traditional methods. Zhang and others. (2020) used deep learning techniques, specifically convolutional neural networks (CNN), to estimate porosity and permeability from seismic data, highlighting the potential of CNN to capture complex subsurface variations. Bhattacharya and Solomatine (2021) comprehensively reviewed various machine learning models used in the oil and gas industry and highlighted the effectiveness of ensemble methods such as random forests and gradient boosting in improving prediction reliability. Lee and Mosser (2022) introduced a hybrid approach combining genetic algorithms with neural networks to optimize the prediction of petrophysical properties, which achieved significant improvements in computational efficiency and prediction accuracy. Zhao and others. (2023) developed a framework for real-time prediction of petrophysical parameters using recurrent neural networks (RNN) that facilitate dynamic reservoir management.