Modelagem da pressão de fundo de poço em sistemas de escoamento multifásico: uma abordagem utilizando modelos de redes neurais polinomiais

Abstract

In the complex oil and gas extraction process, there are many important indicators that must be monitored to maintain the efficiency and safety of the wells, the environment and especially the professionals involved in the operation. One of these indicators is Flowing Bottom Hole Pressure (FBHP). This indicator directly influences production control and well integrity, being a crucial parameter to avoid operational failures and minimize risks. Both low FBHP (which reduces production) and high FBHP (which entails various risks, including the risk of explosion) must be a constant focus of attention. In this sense, computational models have been developed to predict FBHP, providing additional or alternative forms to traditional measurements of this indicator. This dissertation explores the GMDH (Group Method of Data Handling) technique in FBHP predictive modeling. GMDH refers to a family of algorithms that is still little explored, especially in the oil and gas area. This research aims to fill this gap, evaluating the technique as a possible alternative in FBHP modeling. A comparative analysis is made between four different types of GMDH algorithms, pointing out advantages and limitations in terms of precision, generating interpretable models relating to the algorithms in their best performance configuration.