A Report Generation System for Well Logging Industry Based on Large Language Model and Retrieval Augmented Generation

Abstract In the well logging industry, interpretation reports are critical for evaluating subsurface geological conditions, yet their preparation is often time-consuming and inconsistent due to reliance on human experts. This paper presents a novel methodology that leverages the generative capabilities of large language models (LLMs) to automate the creation of well logging interpretation reports, significantly improving efficiency. Central to our system is an LLM-based content generation module that sources and composes report sections. To address data privacy and cost concerns, we utilize open-source LLMs deployed locally. Recognizing the limitations of general LLMs in handling field-specific inquiries, retrieval-augmented generation (RAG) framework is employed to integrate well logging domain knowledge, ensuring contextually relevant outputs. Through experimentation, it is found that prompt phrasing significantly influences LLM performance, prompting the use of prompt engineering techniques to optimize results. A blind review of 100 generated reports by well logging experts demonstrates the system's ability to produce complete, clear and coherent reports. These findings highlight the transformative potential of customized LLMs, capable of generating reports in just 10 minutes—compared to 4 hours manually—thereby revolutionizing labor-intensive processes and significantly improving work efficiency in the well logging industry.