Desenvolvimento de solução CRISP-DM para classificação do evento prisão de coluna no processo de perfuração de poços offshore

Abstract

Over the years, the Exploration and Production (E&P) industry, in order to improve productivity and reduce costs, has sought the use of digital technologies to mitigate the effects of seasonal oil prices, the shortage of professionals, increased competition and to promote safer operations. Today the digital transformation initiativereferred to as the new industrial revolutionhas been driving the use of disruptive technologies and provoking cultural changes, both in society and industry. At the heart of this revolution is the exponential growth and data availability. In this scenario, a change is needed in the way these data are collected, stored, analyzed and accessed to support organizations intelligence and decision-making cycles. This dissertation addresses the process of knowledge discovery in data, through the development of a computational solution with support of the methodology CRISP-DM (Cross Industry Standard process for data Mining) for analysis of anomalies in the process of Drilling of wells in ultra-deep waters. Well drilling is one of the stages that most demand financial resources. Thus, understanding and anticipating problems, evaluating their causes and planning solutions are necessary for global cost control, in order to ensure the well integrity and stability, avoiding non-productive times. This research was delimited to the stuck pipe event and the use of the logistic regression method in the development of the modeling step. The data used were made available by the Brazilian National Agency of Petroleum, Natural Gas and Biofuels (ANP), through a development and innovation research Project (PD&I), containing a set of real data for drilling oil wells of the Pre-salt polygon on the Brazilian coast. At the end of the evaluation process, it was possible to analyze the applicability of the chosen model and the results obtained with the business perspective, that is, that the results are adequate to support the decision-making of the organization. The results obtained to exemplify the model demonstrate an accuracy of 89% with a rate of 99%. Another important result of the work is the contribution to professionals and companies that need to apply methods of data science in similar cases or with other characteristics. Some limitations with low data quality and sample size are also highlighted during the knowledge discovery process. Additionally, programming language codes were included, used for understanding and processing data and generating results. These can serve as an initial version for other analyses.