Modelos de apoio à prevenção de falhas em plantas de petróleo e gás baseados em técnicas de ciência de dados e aprendizado de máquina

Abstract

The need to meet growing energy demand drives the oil industry to explore new sources, notably exploration in deep sea environments. This study is dedicated to the development of a tool for preventing failures in installations intended for oil and gas extraction. The approach adopted consists of the analysis of real data sets, obtained through sensors distributed on multiple oil platforms, accessible in the public database called 3W. Several techniques were applied in data analysis, including: Extreme Gradient Boosting, Random Forest, Extra Trees Classifier, Light Gradient Boosting Machine, Decision Tree Classifier, Linear Discriminant Analysis, Ridge Classifier, Knearest Neighbors, Quadratic Discriminant Analysis, Ada Boost Classifier , Dummy classifier, Naive Bayes, Multi-Layer Perceptron Classifier, Logistic Regression, Support Vector Machine and a Long Short-Term Memory neural network. The results indicated that the Extreme Gradient Boosting model achieved the best performance, achieving an F1 score of 0.99. The project implementation was carried out using the Python language and the pycaret, tensorflow, tsfresh and scikit-learn libraries. The main objective of this model is to provide support to operators in decision-making, contributing to the prevention of failures and optimization of operational performance in oil and gas extraction activities.