The Prediction of the Rate of Penetration Using Artificial Neural Networks and applying the Fuzzy AHP Method for Weighting Input Parameters in One of the Western Oilfields of Iran

Determination of the rate of penetration is one of the most important factors in oil industries. Generally, two methods have been proposed for modeling the rate of penetration which include physical and artificial neural networks methods. Artificial neural networks can be used accurately in order to predict the rate of penetration in which the prediction of the rate of penetration does not include experimental coefficients and bit specifications. Furthermore, in this method, the rate of penetration only depends on the input data. In this paper, the rate of penetration using almost 2000 daily drilling reports and rock mechanical properties was applied. These data were reduced to approximately 1800 data according to data preprocessing. The rate of penetration was modeled by two artificial neural networks including a Multi-Layer Perceptron and an Elman with a hidden layer. After preprocessing the input data and sensitivity analysis of the number of neurons in the hidden layer, seven neurons were chosen as the optimized number of neurons in the Multi-Layer Perceptron with the correlation and mean square error of 77.1% and 1.31, respectively. Also, the Elman neural network showed the correlation and mean square error of 77.6% and 1.33, respectively. Thereafter, the fuzzy AHP method was applied for imposing weights, gaining by expert comments, on the input data resulted in improvements of the artificial neural networks. The results of this investigation have shown insignificant superiority of the Multi-Layer Perceptron neural network for prediction of the rate of penetration comparing to the Elman neural networks. Therefore, the proposed weighted Multi-Layer Perceptron neural network models the rate of penetration accurately and appropriately using available data.