3D rock strength prediction by an innovative approach that integrates geostatistics with machine deep learning models

Abstract This study aims to investigate the limitations of geostatistical prediction models outside the observed data range for estimating rock strength in nonreservoir formations in large geological fields with limited wireline data. To address this gap, this method explores alternative approaches to estimate rock strength using minimum data. A novel 3D rock strength prediction model that integrates geostatistic with deep learning algorithms is proposed. Initially, the deep learning model is trained using the available dataset to capture the complex nonlinear relationships within the data. The developed model is used to increase the dataset size by focusing on nearby data points to mitigate geological variability. geostatistic methods are then applied to establish spatial correlations of rock strength across an extended range compared with those of the actual dataset. The results reveal marked improvements in both the prediction range and spatial resolution of rock strength through the proposed methodology. The developed deep learning models achieved coefficient of determination values ranging from 0.9 to 0.99, demonstrating excellent predictive capability. Cross-validation confirms the model effectively captures local variations. The prediction range in the field expanded by 250% compared to the initial dataset, successfully addressing areas that previously exhibited flat readings when the model was applied to the initial data. This study advances petroleum industry knowledge by integrating deep learning and geostatistical methods to overcome rock strength prediction limitations in nonreservoir formations. The novel 3D model enhances the prediction range and spatial resolution, addresses data gaps and enables better decision-making for areas with limited wireline data.