Comparison of Perceptron and Recursive Artificial Neural Networks In Recognition of Oil Reserves

Mehrafarin Moghimihanjani¹ Farhad Iraji²

1) Chemical & Petroleum Engineering Department, Sharif University of Technology, Tehran, Iran -
2) Chemical & Petroleum Engineering Department, Sharif University of Technology, Tehran, Iran -

IOI Code : XCAZ-EFGEK
Publication : 2nd. International Congress on science & Engineering - paris(parisconf.com)
Abstract :
    The structure of hydrocarbon reservoirs is so complex and heterogeneous that is impossible to identify them directly. Therefore indirect methods of recognizing oil reservoirs such as well tests have become very popular. Identifying the reservoir and boundary model using pressure derivative charts is one of the most important steps in identifying a reservoir. In this study, artificial neural networks such as perceptron and recurrent were used to identify the homogeneous reservoir model and double porosity with four infinite outer boundaries, constant pressure, completely closed, and single- fault without stream. The optimal structure of the proposed networks is using the methods of trial and error and minimizing the mean relative error of the specified training and experimental data. The accuracy of the proposed neural network classification has been considered and compared using simulation and field data. The results showed that the neural network-based model was able to identify the actual model of the reservoir and boundary from the well data. The recurrent neural network has shown that it is more qualified and accurate than the perceptron network and is more capable of classifying experimental well data. These networks have a total accuracy of 98.39%, while the perceptron network provides an accuracy of 95.83%.

Keywords : well test pressure derivation artificial neural networks reservoir model and boundary