| Pressure/Volume/Temperature (PVT) properties are very important in the reservoir engineering computations. There are many empirical approaches for predicting various PVT properties using linear/non-linear multiple regression or graphical techniques. Recently, researchers utilized artificial neural networks (ANNs) to develop more accurate PVT correlations. These achievements of ANN open the door to both machine learning and softcomputing techniques to play a major rule in petroleum, oil, and gas industry. Unfortunately, the developed ANN correlations have some limitations as they were originally developed for certain ranges of reservoir fluid characteristics and geographical area with similar fluid compositions. Accuracy of such correlations is often limited and global correlations are usually less accurate compared to local correlations.;This thesis proposes functional network (FNs) as a novel approach for predicting the PVT properties in reservoir characterization. FN can be considered as a powerful generalization of the standard neural networks, which dealt with general neuron functions instead of sigmoid-like ones. The functions associated with the neurons are not fixed but are learnt from the data. To demonstrate the usefulness of the functional networks technique in petroleum engineering area, we describe both the steps and the use of functional networks for predicting the PVT properties of oil field Brines. In this work, we are going to utilize two types of functional network models, namely, separable and associativity functional networks in forecasting the PVT properties in reservoir characterization. A comparative study will be carried out to compare the performance of this new computational intelligence scheme with the most popular data mining and machine learning forecasting models, such as, feedforward multilayer perceptron neural networks, multiple linear/nonlinear regression, and the empirical correlations algorithms. The results show that the performance of functional networks outperforms all of the existing approaches. In addition, it is more accurate, easier and quicker to train with no over-fitting problem. Furthermore, it dealt with unknown neuron multiargument neuron functions that can be determined from the data at hand. Both results and accurate performance encourage us to say that functional networks can be powerful computational intelligence tools for different applications in oil and gas industry, such as, 3D seismic data, well logs, and history matching, and hydrocarbonate reservoir characteristics. Therefore, we hope that the petroleum engineering scientists use functional networks and consider them as a valuable computational intelligence alternative forecasting approach. |
