Predição da temperatura de desaparecimento de parafina através de redes neurais artificiais

Abstract

Wax deposition in pipelines is a recurring problem in the oil industry that has gained even more relevance with the growth of offshore production in deep environments. Therefore, several studies have been conducted aiming at a better understanding of such phenomenon, as well as to predict wax disappearance temperature (WDT), which represents the true solid-liquid equilibrium point. Information such as this can be useful to support the decision-making process related to pipelines and production units. However, the study of wax deposition is strongly dependent on the performance of experiments, which are often costly and may make the analysis unfeasible. As an alternative to experimental and thermodynamic methods, a model was developed based on machine learning techniques to predict this phenomenon. In this study, an artificial neural network (ANN) was proposed in order to test if there is a model capable of predicting the paraffin disappearance temperature (WDT) using pressure and molar mass of crude oil as input variables. The ANN was trained using different architectures in order to optimize it in relation to the number of neurons in the hidden layer. Results showed that the architecture with 3 neurons in the hidden layer was able to predict the paraffin disappearance temperature with mean square error below (MSE) of 1% and correlation coefficient (R2) of 0.94. The obtained results showed that the proposed artificial neural network is generalist and able to accurately predict the system, without the interference of overfitting and underfitting phenomena. The data obtained made it possible to carry out the sensitivity analysis, in which the pressure was the most decisive independent variable in the process, for the analyzed conditions.