Large Language Model-Based Workflow for Optimizing Offset Well Data Analysis and Generating Well Design Risk Profiles

Abstract Today's workflows for designing and approving new wells depend on modern well engineering practices that integrate input from multiple disciplines like subsurface, engineering, and drilling. Effective execution and planning within these workflows involve complex data analysis including data from historical events from nearby- and offset-wells. Accessing and preparing all this data requires extensive data reviews, that when relying on the historically mainly manual processes and tools, can be both time-consuming and resource intensive. To achieve a step change in efficiency there is a need for new, highly automated engineering solutions to streamline and enhance the end-to-end workflow efficiency. To tackle this challenge, we present an approach that employs new Large Language Models (LLMs) to effectively automate the review process for well engineers. Our LLM-based solution enables well engineers to "interact" with the tools and workflows in novel ways that quickly lets them find and review data from offset wells and historical events. The offset well data was prepared to integrate seamlessly with the LLM-based solution. Retrieval Augmented Generation (RAG) and embedding patterns were incorporated to effectively identify and analyze relevant events within the new well design context. The LLM-enabled tool facilitates a workflow that still relies on proven tools to perform risk analysis of library data, but the UI experience is completely transformed. In conjunction with event analysis, this provides a much more efficient way? to identify risks related to the new well design. The workflow not only streamlines the review process but also enhances the quality and comprehensiveness of the risk assessments. In this paper, we discuss the workflow facilitated by the LLM-based assistant tool, including the design considerations for the tool. We cover both the design of data engineering and the development of data analysis and LLM patterns. Considerations of system complexity, implementation cost, and operation are also discussed. This work serves as a guideline for the implementation of other LLM-based solutions and tools for well engineering challenges and beyond.