Modelagem matemática do comportamento de fluidos de perfuração com o uso de aprendizado de máquinas

Abstract

Predicting the rheological behavior of drilling fluid is essential for oil well drilling, as the operation’s success depends on the functions the fluid must perform. This work aimed to develop a Machine Learning algorithm capable of representing the rheological properties of drilling fluids under various conditions. The development platform used was Google Colab Navigator, with Python being employed for data analysis, exploratory computation, and visualization. The dataset was provided by studies conducted at the LabReo rheology laboratory, derived from the rheological characterization of water-based and olefin-based drilling fluids. The data was prepared, preprocessed, and normalized. An artificial neural network model was developed and trained using hyperparameters defined through comparisons with performance metrics. The flow curves of the fluids were plotted and compared, assessing the accuracy of the predictions made by the developed model against experimental values. Linear regression and Power Law rheological models were implemented for comparison with the neural network model. The optimization of hyperparameters played a crucial role in the obtained results. The best hyperparameter configuration included five layers with a neuron sequence of 16, 32, 64, 32, 16, a learning rate of 0.001, and ReLU activation function. The number of training epochs influenced the accuracy of shear stress predictions and using 300 epochs improved accuracy without excessively prolonging processing time. Striking the right balance between epochs and accuracy is essential for more precise results. The developed neural network model demonstrated high accuracy in predicting flow curves, making it a powerful tool for forecasting the rheological behavior of drilling fluids under different conditions.