A New Model for Predicting Surface Pump Pressure of Drilling Rig Using Artificial Neural Network

Abstract

Machine learning and artificial intelligence are recently used in many engineering sectors. Artificial neural network (ANN) has been widely used in oil and gas to predict many important parameters. This work uses ANN to predict the required surface pump pressure at the surface, considering the impact of different drilling parameters. These parameters are: depth, rate of penetration (ROP), weight on bit (WOB), rotation per minute (RPM), stroke per minute (SPM), mud weight, and mud flow rate. ANN models were built using two layers, and both hyperbolic Tanh and Log sigmoid transfer functions were used to predict the model’s validity. Around 2020 data values were used to test, train and validate model prediction. Sensitivity analysis used 2, 4, 8, and 10 neurons for each transfer function (Log sigmoid and hyperbolic Tanh). Results indicated that the prediction for the eight nodes Tanh model best matches the overall data available for the test. For instance, a 99.67% R for training, 99.45% test, 98.57% validation, and 99.47% overall data set were obtained. On the other hand, using a Log model with ten nodes offered the best data set matching for the same data tested above. Results show that test data converged 99.58 with the model prediction method, while 99.52 and 98.95 were obtained for training and validation, respectively. Therefore, we suggest a new model based on the Log model to predict surface pump pressure. This model would be beneficial for predicting the required number and size of pumps at any drilling site.