EURO 2025 Leeds
Abstract Submission

2924. Improving Decommissioning of Offshore Platforms using Generative AI: An application of Large Language Models in the Oil and Gas industry

Invited abstract in session TD-38: Foundation Models and Optimization, stream Data Science meets Optimization.

Tuesday, 14:30-16:00
Room: Michael Sadler LG19

Authors (first author is the speaker)

1. Leonardo S. L. Bastos
Industrial Engineering, Pontifical Catholic University of Rio de Janeiro
2. Pedro Hamacher
PUC-RIO
3. Pedro Lobato
4. Gabriela Ribas Klein
Tecgraf, PUC-Rio
5. Luidgi de Jesus da Silva Colimerio
Informática, Pontifícia Universidade Católica do Rio de Janeiro
6. Adriano Simões
Instituto Tecgraf, Pontifícia Universidade Católica do Rio de Janeiro
7. ALEXANDER FERNANDES
Descommissioning, Petrobras
8. Marcos Vinicius Marques da Silva
Tecgraf, PUC-Rio
9. Vitor Sabbagh
Petrobras

Abstract

When offshore oil extraction platforms are no longer economically viable, they enter the decommissioning phase (DP), during which the structures are removed from the sea, a process that involves several activities and whose proper execution is crucial to reduce impacts on the marine environment.
During DP planning, it is required to describe the platform’s operational history, requiring labor-intensive activities to search and summarize decades of reports, which may delay the whole process and is prone to inconsistencies. Recently, the use of language-based Generative Artificial Intelligence (GAI) such as the Large Language Models (LLMs) has shown high potential in improving processes, especially regarding the processing of text-based data.
Hence, this work proposes an LLM-based system to process and generate summarized information for the DP in a large Oil and Gas company.
A Retrieval Augmented Generation (RAG) framework was developed, consisting of two stages: searching the report database with semantic search and filters to identify the documents relevant to that oil well and using them to draft the final document with an LLM.
Computational experiments were conducted to optimize the framework’s efficiency, using metrics designed for RAG evaluation to calculate its ideal hyperparameters. This work connects the domains of Oil & Gas and Generative Artificial Intelligence with a practical use case, automating processes for productivity optimization and cost reduction.

Keywords

Status: accepted

Back to the list of papers