A Multi-Scale Cross-Correlation Based Method Enables Automatic Well Log Depth Matching

Abstract Depth matching is a critical element in integrated well log interpretation. However, it remains a challenge for the industry that depth alignment must be done manually or semi-manually. With increased interest in data science leveraging logs from hundreds of wells, there is a strong desire for automating the depth matching process for the benefit of efficiency and avoiding human bias. We developed a solution to automate log depth matching based on a multi-scale cross-correlation method. The solution adopts a philosophy of matching outstanding features while linearly adjusting other depths, akin to human logic in manual depth shift. It utilizes log-squaring and perceptually important point algorithms to identify outstanding features from the noisy well log. To achieve robust and automatic depth matching, a multi-scale cross-correlation based optimization process is adopted to align the identified outstanding features. A fast pre-shift step eliminates the need for large search windows during the optimization process, which is crucial for efficiency and full automation. This approach accommodates pairs of well logs that have reasonable correlation, either positive or negative. The developed solution was first applied to match depth in gamma-ray (GR) logs. Gamma-ray logs are widely accepted by the industry for depth synchronization between different logging passes as an indicator of lithological changes. The basic process is to build a depth mapping table from a reference gamma-ray log and a target gamma-ray log, which are acquired from two separate logging passes, using the developed solution. Then, apply the obtained depth mapping table to other well logs in the same logging pass to synchronize all logs and make them ready for the downstream processing. Excellent results have been demonstrated through field gamma-ray logs, for both wireline to wireline and wireline to logging-while-drilling applications. The solution was then applied to match depths for other well log types, including density to sonic slowness, and slowness logs from different sonic modal processing, all demonstrating excellent performance. Benchmarking against previous commercial solutions shows significant improvement without the need for human intervention, proving a fully automated solution for well log depth matching. To the best of our knowledge, this is one of the first algorithmic solutions in open literature aiming to automate well log depth matching based on non-machine learning methods. This solution combines a two-step shifting strategy, multi-scale cross-correlation optimization, and a feature selection algorithm to form a unique, fully automated solution that addresses the well log depth matching problem, which has been a long-standing challenge in the industry.