Geology-Informed Hybrid AI For Automated Well Log Marker Propagation

Subsurface geological and petrophysical interpretation benefits from precise formation depth information. However, such information is often lacking or of low quality due to unprecise marker depth reporting. Wellbore data, including well logs, are a valuable source of information to propagate markers. We leverage this data in the novel automated approaches we introduce for marker propagation. We first introduce a dynamic time warping log template matching algorithm based on well log signature clustering and neighboring log template identification. We then introduce a novel geology-informed hybrid AI method that leverages a deep learning model, combined with a multi-log change-point detection algorithm, to detect candidate marker depths. Such candidates are then compared against neighboring well markers to optimize the well log similarity across common markers intervals. We evaluate both approaches on a public sample of 103 wells from the Norwegian Offshore Directorate to propagate groups. While the dynamic time warping approach succeeds in propagating markers with clear and distinctive well log signatures, the hybrid AI method achieves a significantly more precise propagation. The first approach provides a 111 feet median error compared to the ground truth marker data, while the hybrid AI achieves a 77.5% lower error with a 24.9 feet median error.