This study has been undertaken in two oil fields (A-Libya, and B-Libya) in Sirte basin located in Libya. Nubian sandstone Formation is the main reservoir in the studied oil fields. Laboratory resistivity measurements were performed at Libyan Petroleum institute (LPI). The majority of wells, however, are logged and the use of wireline log data in conjunction with some core data has been proposed as a rapid, cheap, and alternative to predict some special core analysis (SCAL) parameters instead of collecting extensive core or performing SCAL measurements in all wells. Neural network predictors are potentially very useful in the present study due to the limited SCAL data for the studied wells. In this work some of SCAL parameters were predicted using neural networks based on different combinations of wireline logs. The procedure firstly involved training the neural network predictors using data in training well A-02. These predictors were then applied to an adjacent well A-01 in the same oil field, and to another test well B-01 in a different oil field. The most frequently used type of neural network is a feed forward neural network using a back-propagation learning algorithm, due to its popularity and simplicity.
Some good neural network SCAL parameter predictors for Rt, and RI were generated using different combinations of standard wireline logs in the training well A-02. The best predictors were produced using the dataset from the entire 478 ft cored interval of the training well and all 7 available wireline logs. Predictors trained on data at 1.0 ft depth spacing appeared to be better in the training well. However, the prediction of resistivity parameters in an adjacent well and a further test well in a different oil field gave slightly better results in general for predictors trained on data at 0.5ft depth spacing rather than at 1.0 ft depth spacing.