|dc.description.abstract||One of the main objectives of this work was to investigate the applicability and accuracy of artificial neural networks for estimating special core analysis (SCAL) parameters from minimal core training data and wireline logs. The SCAL data was obtained from measurements on core plugs undertaken at the Libyan Petroleum Institute (L.P.I). Previous neural network studies have attempted to predict routine core analysis parameters, such as permeability, but not SCAL parameters such as true formation resistivity (Rt), resistivity index (RI), water saturation (Sw), saturation exponent (n) and Amott-Harvey Wettability Index (IA/H). Different combinations of wireline logs were used to train a variety of neural network predictors. Some of the predictors were trained using a large dataset from the entire cored interval of the training well. Other genetically focused neural network (GFNN) predictors were trained just from one short representative genetic unit (RGU) in the training well. The predictors were then tested in an adjacent well in the same oil field and also in another well in a different oil field. Significantly the performance of the GFNN predictors was as good (and in most cases better) than the predictors trained on the much larger dataset. This demonstrated the useful of the GFNN approach, which is very cost effective in terms of the minimal core that is required, and the reduced computer processing time. Moreover, this is the first time that these GFNN predictors have been used to predict SCAL parameters in the studied area, the Nubian Sandstone Formation in North Africa. These neural network predictors are particularly useful in this area due to the limited amount of SCAL data that is currently available.
Quantitative statistical measures of heterogeneity were also examined on the reservoir samples, followed by a comparative analysis of hydraulic units (HUs) with a newer approach of global hydraulic elements (GHEs) to characterize the reservoir units in the studied area. The GHEs were then applied to select minimal representative core training data to train the genetically focused neural networks (GFNNs) to predict the SCAL parameters.
The thesis also describes the factors affecting SCAL resistivity parameters. Laboratory measurements on the Nubian Sandstone reservoir rock samples showed changes in the formation resistivity factor (F) and cementation exponent (m) between ambient conditions and at overburden pressures. Changes were also observed in the saturation exponent (n) before and after wettability measurement. The experimental results also showed that there was a good relation between resistivity and the type of pore system which is consistent with study result from Swanson (1985) confirming earlier work.||en_US