Reservoir characterization through comprehensive modeling of elastic logs prediction in heterogeneous rocks using unsupervised clustering and class-based ensemble machine learning

Abstract

Geophysical reservoir characterization is a significant task in the oil and gas industry and elastic logs prediction of subsurface formations is a fundamental aspect of this process. However, elastic log prediction in a high-dimensional and complex geological environment, such as the Lower Indus Basin Pakistan, poses a significant challenge where traditional empirical methods often fail to provide competitively accurate results. Therefore, this study proposes a novel machine learning approach that combines unsupervised clustering (K-means) and ensemble-based machine learning (random forest) to improve prediction accuracy. By clustering data based on statistical similarity and ensemble algorithms to each cluster, the methodology addresses the challenges of sonic log prediction in the Lower Indus Basin (Pakistan). This approach was evaluated using real-world data, outperforming several baseline methods with a root mean square error of 98% accuracy. Its effectiveness to predict elastic log makes it a valuable tool in reservoir characterization, earthquake analysis, and geothermal energy exploration. Overall, combining this methodology with other techniques can enhance seismic data analysis and enable better decision-making in the oil and gas industry. This novel approach presents an effective solution for predicting sonic log in the Lower Indus Basin and contributes to advancements in geophysical reservoir characterization.