A Case Study of Artificial Neural Networks Usage for Rate of Penetration Prediction with Offset Wells Data

Resumo

The oil and gas industry is constantly motivated to implement strategies focusing on cost management and operational optimization to maximize productivity. Rate of Penetration (ROP) serves as a key performance metric, reflecting the speed at which a drilling bit penetrates subsurface formations. Increasing ROP can minimize costs by reducing drilling time. ROP prediction models aid drilling optimization plans since accurate models are required to find ideal controllable parameters to increase ROP using optimization algorithms. Drilling characteristics that significantly influence ROP include bit type, formation properties, and operation parameters, such as Weight on Bit (WOB) and Rotations per Minute (RPM). The number of significant parameters makes it difficult to develop analytical expressions for predicting ROP. The present work investigates the applicability of Artificial Neural Networks (ANN) for ROP prediction using public data from three wells extracted from the Volve oil field in the North Sea. The adopted strategy is to learn from offset wells, which means using data from two wells to predict the ROP for the third. This approach helps to address a practical scenario where historical data are used to make predictions for a new well in a close region. This study uses MLP (Multilayer Perceptron) networks with WOB, RPM, Torque, Fluid Flow rate, and Delta-T Compressional as features, and the training process is conducted using the GridSearchCV method for hyperparameter tuning, comparing different model architectures to improve results. To assess the final performance of the model, metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are applied to the tested well. The results show that the developed model can capture ROP patterns without requiring any data stratification of the two wells regarding lithology information.