Inter-Well Similarity Analysis as a Key Pre-Processing Step in Stratigraphy Interpretation Workflows

Abstract

Abstract High-resolution and high-quality stratigraphic markers are fundamental to geological modelling and operations. Stratigraphic interpretation workflows, however, are time-consuming due to the processing of massive amounts of data collected from geological surveys. This paper presents highlights of the importance of inter-well similarity analysis on the preprocessing stage for stratigraphic interpretation. The proposed methodology involves the use of common time series similarity metrics for quantitative estimation of pairwise similarity between wells to select appropriate ones for further interpretation. Experiments were conducted on the data collected from Groningen Gas field and petroleum basin in the Middle East. The impact of data selection on the final results was estimated based on the prediction quality metrics and computational time between baseline models (XGBoost classifier and simple Feedforward Neural Network) trained on the whole set of wells and models trained only on selected wells. The integration of similarity-based wells selection step allowed us to improve computational time of models, while preserving or even outperforming prediction quality. With the results of our experiments, we demonstrated the importance of the usage of only representative data for stratigraphic predictions and custom-made models’ creation for the target well. The proposed workflow can be beneficial for traditional manual correlations as well.