An ROP Predictive Model in Nile Delta Area Using Artificial Neural Networks

Abstract

Abstract Bit performance is a key factor to improve drilling performance and reduce drilling costs; however, factors controlling bit performance are many and the interactions between these factors are very complicated. This research provides a tool that represents this complex physical phenomenon using artificial neural networks (ANN) to predict the rate of penetration (ROP) to optimize bit selection. Lithology changes, drilling parameters data, and bit data were the inputs of our model. Field datasets were collected from six different wells in the Nile Delta area represented by more than 12,000 records; these records were subjected to statistical analysis to exclude the data outliers and bad quality records. The cleansed datasets (more than 7,000 records) were normalized as a preprocessing step prior to building the ANN model. Many exploratory models were developed prior to building the final model. These exploratory models indicated that a back-propagation feed-forward ANN is the proper ANN type that will produce acceptable representation of the physical process with reasonable predicted values of ROP. The developed ANN model was trained, validated, and tested to predict ROP using the cleansed data set. The correlation coefficient was over 0.90 for the training data. For testing data, the correlation coefficient ranged from 0.88 to 0.90. Statistical analysis of the model outputs comparing the actual measured values shows a low average absolute percent error, low standard deviation, low maximum error, and low root mean square error. When datasets from outside the selected research area were used for prediction, the model introduced an unsatisfactory result of ROP predictions with a correlation factor less than 0.5; which means that the model should be retrained using data records from the area of concern before using the model for predicting ROP. The newly developed model can be used only within the range of the training data used; therefore, care should be taken if other data beyond this limit is utilized.