Abordagem evolutiva de aprendizado de máquina para caracterização litológica de poços de exploração de petróleo.

Abstract

The total organic carbon of rock samples is the main detailed and qualitative measure of the organic matter present in a basin. It is often manually calculated by studying source samples. This approach, however, requires time and money because it relies on withdrawals from various well intervals in source rocks. As a result, research has been carried out to assist in this endeavor. Machine learning algorithms are an alternative to providing estimates of total organic carbon based on data well records and stratigraphic analysis. In this scenario, the current one suggests an estimate of total organic carbon that is thought using the machine learning approach. To add flexibility to the model, the selection of the cross parameters was performed using a standard heuristic approach with validation. This computational technique allows the identification of models with greater generalization potential. Convolutional Neural Networks (CNN), Extreme Learning Machine, Elastic Net Linear Model and Extreme Gradient Boost were used. The suggested approach was validated using samples from some sedimentary basins. In research capacity, the RNC technique excels approaches, demonstrating in helping the prediction of total organic carbon.