Prediction of Flowing Bottomhole Pressure (FBHP) through Optimized Machine Learning Techniques.

Resumo

Flowing Bottomhole Pressure (FBHP) prediction is important for optimizing oil and gas production, as it directly impacts operational efficiency and hydrocarbon recovery. In this case, physical and numerical models are widely used, but often face some limitations due to the complexity of the reservoirs and the nonlinearity of the data. Therefore, the importance of FBHP prediction justifies the continuous focus on research and development, aiming at the efficiency, safety and long-term sustainability of the oil and gas industry. To accurately predict downhole pressure, it is crucial to adjust operating parameters such as flow rate and surface pressure, maximizing hydrocarbon recovery and reducing operating costs. Given these limitations, machine learning techniques emerge as viable alternatives, capable of dealing with complex and non-linear data, offering greater accuracy. The present work proposes the use of four machine learning algorithms: Elastic Net, Extreme Learning Machine, Support Vector Machine and Random Forest, and to find the optimal parameters of these methods the application of the metaheuristics: Gray Wolves, Particle Examination, Bee Colony and Differential Evolution, in order to improve the prediction of FBHP. The methodology includes data pre-processing, normalization, feature selection, model training with hyperparametric optimization and cross-validation to ensure effectiveness. Therefore, it is expected to achieve greater model accuracy and performance, with reduced overfitting and greater generalization capacity. This involves advanced optimization, improved computational intelligence, and more effective practical applications, in order to obtain high-performance models, with predictive capacity and computational efficiency.