Artificial neural networks-based correlation for evaluating the rate of penetration in a vertical carbonate formation for an entire oil field

The rate of penetration (ROP) is a critical factor affecting the process of oil well drilling optimization and the total drilling cost. This work introduces an empirical correlation extracted from the learned artificial neural networks (ANN) to assess the ROP across a vertical carbonate formation from five surface drilling parameters measurable through real-time sensors. The ANN was built based on real 220 datasets obtained from eight wells. The data from five of these wells were used to train the ANN model. Several sensitivity analyses were conducted on the model's parameters to achieve the best combination of these parameters. To enable real-time assessment of the ROP, a correlation from the leaned ANN model was extracted, which was tested on 92 datasets from the same training wells while unseen datasets from another three wells were used for validating the empirical correlation. The results showed that the ANN was effectively predicted the ROP with an average absolute percentage error (AAPE) of only 4.34% for the training data. Using the developed equation, the ROP was assessed for the validation data with an average AAPE of 6.75%.