Rate of penetration prediction using self-adaptive differential evolution-artificial neural network

The rate of penetration (ROP) should be optimized during the drilling operation to avoid many problems such as cutting accumulation, pipe sticking, twist off, and to reduce the nonproductive time (NPT). ROP depends on many variables such as; drilling parameters, fluid properties, and formation strength. Previous models of ROP did not include the effect of the change in rock mechanics on the ROP prediction. Several models were developed based on a regression that is limited to the rock type and drilling conditions. The objective of this paper is to apply the self-adaptive differential evolution technique (SaDE) to optimize the artificial neural network (ANN) variable parameters such as; training function, a number of hidden layers, transferring function and number of neurons in each layer. The optimized SaDE-ANN model will be used to predict the ROP as a function of torque (T), weight on bit (WOB), stand pipe pressure (P), flow rate (Q), the drilling fluid density (D), uniaxial compressive strength (UCS), plastic viscosity (PV), and RPM. The obtained results showed that ROP has a strong function of the drilling parameters; RPM, WOB, T, and horse power (HP). While ROP is a moderate function of UCS. The optimized ANN structure based on SaDE is 5-30-1; where the input layer consists of 5 neurons representing the input parameters; RPM, WOB/D, T/UCS, D/PV, and HP. The hidden layer consists of 30 neurons and the output layer contains one neuron representing the output predicted ROP. The ratio of the training and testing data was 0.6 and 0.4, respectively. The best training function was Bayesian Regularization backpropagation (trainbr) and the best transforming function was Logarithmic sigmoid (logsig). The high accuracy of the developed model confirmed the importance of compiling the drilling fluid properties with the drilling parameters.