Optimum Drill Bit Selection for Horizontal Wells Utilizing Neural Network Model

Abstract

An artificial neural network is a computational system of networks and consists of a simple element called neurons or nodes that simulates human intellectual processes, like learning, training, making decisions, and solving problems with multiple variables and a high number of hypotheses. Utilizing rate of penetration data and other drilling parameters, an artificial neural network model was developed to select the optimum drilling bits in the current study. The suggested model was programmed using MATLAB. The neural network adopted a two-layer structure with a sigmoid transfer function employed in the hidden layer and the output layer. The hidden layer consists of three neurons, whereas the output layer holds one neuron. The data on the rate of penetration, other drilling parameters, and formation characteristics were collected from horizontal wells of one of the Iraqi fields. Preprocessing involved cleaning, normalization, and handling of missing values to ensure data quality. The neural network was trained with a backpropagation algorithm and validated on a separate dataset. The performance evaluation results using the R-squared coefficient demonstrated good agreement between actual and predicted rate of penetration values, with an R-squared coefficient value of 0.9982 when using nine neurons, which indicates that improved prediction accuracy is a result of increasing the number of neurons in the hidden layer.