Enhancing AI Litho-Fluid and Virtual Sonic Models for Geological and Geophysical Well Real-Time Characterization

This paper presents advancements in AI-driven Litho-Fluid Interpretation and Virtual Sonic Log Reconstruction models for real-time geological and geophysical well characterization. Integrated within a real-time digital ecosystem, these models enhance operational decision-making during drilling by automating lithology and fluid identification and sonic log generation. The Litho-Fluid model uses a facies-based expert system with six one-class classifiers, improving accuracy over previous approaches. It operates in three tiers—Drilling, Simple, and Complete—based on data availability. The Virtual Sonic model reconstructs sonic logs using while-drilling data, addressing compaction effects through depth-based inputs like TVDBML and DT-Vel. Case studies show improved carbonate interpretation and accurate sonic log reconstruction, aligning closely with acquired data. These innovations reduce interpretation latency, improve safety, and optimize drilling performance. Future work aims to simplify models by removing depth-based inputs and inferring compaction directly.