Contribution à la caractérisation des réservoirs fissurés du champ de Hassi Messaoud par classement flou, réseaux de neurones artificiels et magnétisme des roches

摘要

Fractured reservoirs form a particular kind of reservoirs, due to porosity and permeability fracture effects. Optimizing exploitation of hydrocarbons in this type of reservoir requires a specific study compared to other conventional reservoirs. Our study consists in using the maximum available data to better characterize its petrophysical characteristics, despite the lack of the sonic log in the studied wells. The log data are used to estimate natural fracture porosity which is considered to be a key parameter to evaluate and model a fractured reservoir. The prediction of this parameter is addressed using fuzzy logic and neural networks. The Hamra quartzites fractured reservoir situated in the southwest of Hassi Messaoud oilfield is studied using data imaging logs and well cores. The combination of different techniques (X-ray diffraction, scanning electron microscope and rock magnetism) is used to better investigate the nature of magnetic minerals in the studied reservoir. The aim of our study is to look for a linear or a non-linear relationship between magnetic susceptibility and petrophysical parameters by applying principal component analysis and neural networks. The results obtained show that the correlation coefficient (R2) between values of the natural fracture porosity estimated by neural network and those calculated by logs is equal to 0.878. The combined techniques used to identify magnetic mineralogy show that pyrrhotite, hematite and magnetite are the main magnetic minerals responsible of the high magnetic susceptibility intervals in Hamra quartzites reservoir. The prediction of magnetic susceptibility is estimated from log data, using fuzzy logic and neural networks. The results found with a neural network composed of 25 neurons in the hidden layer prove a good performance in the test phase, with a mean square error, a mean relative error and a correlation coefficient (R) equal to 0.0142, 0.0743 and 0.907, respectively. These results confirm a non-linear relationship between magnetic susceptibility and these parameters.