Application of advanced computational intelligence to rate of penetration prediction

Rate of penetration (ROP) prediction is a very important aspect in oil and gas industry. Several studies using different methods were applied to predict ROP. This importance stems from cost reducing of drilling projects. The objective of this paper is to compare the traditional multiple regression method with Extreme Learning Machines (ELM) and Radial Basis Function Network (RBF) as applied to predict ROP. ELM and RBF are artificial neural network (ANNs) techniques. ANNs are cellular systems which can acquire, store, and utilize experiential knowledge. For ELM, the activation functions, number of hidden neurons, and number of data points in the training data set are varied to find the best combination. The dataset is composed of seven input parameters. These are depth, bit weight, rotary speed, tooth wear, Reynolds number function, equivalent circulating density (ECD), and pore gradient. Prediction results found in Eren's multiple regression study are used in the comparison. The comparison is made based on field data of two different wells with no correction, then with weight on bit (WOB) vertical correction, and finally with interpolated WOB and rotary speed (RPM) motor correction. The techniques are compared in terms of training time and accuracy, and testing time and accuracy. Different input parameters of ELM and RBF give different results. The decision makers are advised, according to the results of this study, to choose ELM with sigmoidal activation function, training data = 80% and number of hidden neurons = 10 as ROP prediction technique.