Similarity learning for wells based on logging data

One of the crucial steps geological object exploration is interwell correlation. The correlation matches similar parts different wells helping to construct models and assess hydrocarbon reserves. Today, a detailed relies on manual analysis well-logging data: process prone significant time consumption subjectivity. Alternative automation attempts include rule-based, classic machine learning, more recent deep learning methods. However, most approaches are still limited usage inherit cons We propose method based model solve profile similarity estimation. Our takes data as input, constructs well representations uses them provide wells. developed algorithm enables extracting patterns essential characteristics profiles within follow an unsupervised paradigm that allows us utilize large pools logging available in industry does not rely subjective labelling. For testing, we used two open datasets originating New Zealand Norway. data-based has decent quality. example, accuracy our 0.926 compared 0.787 for widespread gradient boosting baseline. selected representation approach also provides high extrapolation capabilities work formations other than those during training start amounts data, experiments show.