Caracterisation Petroliere De Reservoir Siegenien A L’aide Des Methodes Machine Learning, Application Au Champ De - Wadi El Teh - Bassin De Berkine Ouest

Résumé: The characterization of the reservoir is a very important and crucial step in petroleum exploration as well as production and marketing activities, as its results provide a decisive conclusion on the profitability of the reservoir and the efficiency of the exploration and wells. operation. The Siegenien reservoir is one of the most important reservoirs in the Berkine field in Algeria, where many wells are being drilled in order to improve its productivity and extend the production phase in the first place. However, due to the increase in drilling depths and other technical problems such as swelling clays, high pressure HPHT, the unavailability of radioactive sources and the high cost of certain operations, the Log neutron density cannot be recorded and the porosity is therefore calculated from the sonic logs. In many cases, this sonic porosity represents the effective porosity because the reservoir is not fractured. However, if a secondary porosity exists, the porosity obtained from the sonic log will be different and therefore there will be an overestimation of the oil reserves. Additionally, MDT and coring jobs cannot be processed due to the previously mentioned issues. On the other hand, applications ML have experienced huge and successful development in recent years. It is considered to be an advanced solution to many complex problems in many fields, including the field of tank assessment. Therefore, the objective of this thesis is to use ML in order to calculate the porosity and the permeability in the case where the measurement of the neutron density cannot be recorded; The proposed algorithm was trained on the basis of core porosity and core permeability data from four wells of the Berkine West field (Wadi El Teh), then tested and validated in another well where only the sonic log is available. The artificial intelligence algorithm (Machine Learning) should provide more reliable results for the problem described.

Publié dans la revue: