Application of machine learning algorithms for porosity prediction based on well logging data

Abstract

In the oil and gas industry, porosity is a crucial parameter for determining the viability of a reservoir. Traditionally, porosity is measured by collecting core samples from a well, which is a time-consuming and expensive process. This study aims to investigate the effectiveness of two Machine learning (ML) algorithms, including Support Vector regressor (SVR) and Multi-Layer Perceptron (MLP) for porosity prediction. A dataset of well logs, consists of 292 data points collected from Libya field, was utilized to train and test the ML models. The well logs includes core porosity measurement as a target variable and five input geological measurements such as gamma ray (GR), photoelectric (PE), neutron porosity (NPHI), shallow resistivity (RXOZ) and bulk density (RHOZ). The obtained results reveal that the Both MLP and SVM exhibited good performance in predicting porosity, with the SVR model achieving slightly better results. The MLP model yielded a Root Mean Square Error (RMSE) of 1.3729, Average Absolute Percentage Error (AAPE) of 4.6930, and coefficient of determination (R²) of 0.8952. The SVR model achieved an RMSE of 1.52052, AAPE of 4.5082, and R² of 0. 88321. Furthermore, the findings of this study demonstrate the potential of machine learning algorithms, particularly SVR, for accurate porosity prediction using well logging data. This approach offers a more efficient and cost-effective alternative to traditional core analysis methods for reservoir characterization in the oil and gas industry.