Automated neural network optimization for data-driven predictive models: an application to ROP in drilling

The rate of penetration (ROP) holds significant importance in optimizing a drilling process. ROP assists in alleviating concerns in critical scenarios where limited visibility over explorations reduces efficiency, increases non-productive time, and heavily costs operations. In this study, a comprehensive automated data-driven model for ROP is proposed. The model integrates two optimization algorithms: the conjugate gradient algorithm (CG) for training and updating the parameters of the radial basis function neural network (RBF-NN) and the genetic algorithm (GA) for automated hyperparameter optimization (HPO). The proposed model is applied to a development well with six controllable drilling parameters as inputs and the results of comparison show prediction accuracy is improved by at least 50% compared to the RBF-NN. The findings of this study reveal the importance of parameter optimization and automated hyperparameter optimization to provide a more efficient, unbiased, and scalable approach, leading to improved generalization and performance of machine learning (ML) models. The developed model provides a basis for intelligent optimization and control in complex geological drilling processes.