| One of the major objectives of the petroleum industry is to obtain an accurate estimate of initial hydrocarbon in place before investing in development and production. Porosity, permeability and fluid saturation are the key variables for characterizing a reservoir in order to estimate the volume of hydrocarbons and their flow patterns to optimize the production of a field. Many empirical equations are available to transform well log data to predict these properties. Recently, researchers utilized artificial neural networks (ANNs), particularly feed forward back propagation neural networks (FFNN), to develop more accurate predictions. The success of FFNN opens the door to both machine learning and soft-computing techniques to play a major role in the petroleum, oil, and gas industries. Unfortunately, the developed FFNN correlations have some drawbacks, and as a result several improvements have been proposed.;This thesis investigated the suitability of some of the recently proposed advances in neural networks technique including, functional networks (FN), cascaded correlation neural networks, polynomial networks, and general regression neural networks for predicting porosity and water saturation from well logs. Since there is no fully developed software for functional networks, we described both the steps and procedures in developing functional networks to predict these properties. We also compared the performance of these techniques with standard FFNN as well as the empirical correlation models.;Generally, the results show that the performance of General Regression neural networks, Functional networks and Cascaded Correlation networks outperform that of standard neural networks. In addition, General Regression Neural networks are more robust while Functional networks are easier and quicker to train with no over-fitting problem, and more importantly we have more insight into the coefficients of the network. Therefore, we believe that the use of these better techniques will be valuable for Petroleum Engineering scientists. |
