|
Отправлено: 26.04.13 12:27. Заголовок: Improving Performance of a Neural Network Model by Artificial Ant Colony Optimization for Predicting
Middle-East Journal of Scientific Research 13 (9): 1217-1223, 2013 ISSN 1990-9233 © IDOSI Publications, 2013 DOI: 10.5829/idosi.mejsr.2013.13.9.927 Corresponding Author: Amir Hatampour, Faculty of Chemical Engineering, Dashtestan branch of Islamic Azad University, Dashtestan, Iran, Tel: 0098 939 160 5121. 1217 Improving Performance of a Neural Network Model by Artificial Ant Colony Optimization for Predicting Permeability of Petroleum Reservoir Rocks Amir Hatampour, Rasul Razmi and Mohammad Hossein Sedaghat Faculty of Chemical Engineering, Dashtestan Branch of Islamic Azad University, Dashtestan, Iran Abstract: Over recent years, oil and gas exploration has been beloved due to extended needs for petroleum and energy sources. In this way, the capabilities of intelligent techniques are studied by researchers in different fields of petroleum industry and geosciences, whereas it seems that these techniques can improve the prediction accuracy of exploration and production of the hydrocarbon reservoir. The performance indices of the artificial models have proven to be better than the conventional linear and non-linear statistical models. Artificial intelligence models are extremely useful for reservoir characterization, which requires a high-accuracy prediction for a good exploitation of the oil and gas resources. The present paper introduces an optimized neural network (NN) for predicting permeability based on a relatively new swarm intelligent approach, named ant colony optimization. Ant colony optimization (ACO) is a relatively new computational intelligent approach, which was initially inspired by the observation of ants. The number of neurons in hidden layer, weights and bias are optimized in the proposed NN using the ACO. The data of a gas reservoir in the South Pars Gas Field of Iran was used for analyzing the accuracy of optimized NN in a real case study. Finally, to clarify the advantages of the optimized NN, its outcomes were compared with the results of a simple NN model, in which the aforementioned parameters were determined through a try-and-error process. The MSEs of the optimized and simple NN were equal to 7.95 md and 12.84 md, respectively, which correspond to the correlation coefficient (R) of 0.94 and 0.866, respectively.
|