Characterizing reservoir fluid composition is a crucial step in all phases of the exploration and development cycle of any oil field. The petroleum industry has dedicated much work to build up computational techniques to model phase behavior through two approaches using conventional regression methods and the Equation-of-State models. Recently, Artificial Neural Networks (ANN) have been successfully used to solve numerous problems in the petroleum industry. This interest is due to the fact that neural networks generally have large degrees of freedom, thus they can capture the nonlinearity of the process being studied better than regression techniques. In addition, neural networks have the ability to model systems with multiple inputs and outputs.
The objective of this study was to develop a neural network model to predict the molar compositions of heavy oil as functions of the well location, depth, bottomhole temperature, bottomhole pressure, API gravity, and gas gravity. Since composition may change vertically as well as laterally, well location and depth of the samples are included in the training of the neural network. A total of nine pseudocomponents are used as output variables.
The data, on which the network was trained, contained 82 data sets collected from Lower Fars heavy oil reservoir of north Kuwait. Several neural network architectures were investigated to obtain the most accurate model. Results indicate that general regression neural network shows optimum prediction capability for molar compositions as functions of the input variables. In addition, the model was able to successfully predict the molar compositions from inputs that were not seen during the training process. The output of this study is in accordance with the new vision and strategy that Kuwait Oil Company has set toward the exploitation of the unconventional heavy oil resources of Lower Fars reservoir of North Kuwait in which a lot of new wells will be drilled in this area. A development of such a new and innovative prediction model using ANN approach is of great importance in providing the molar composition of Lower Fars heavy oil reservoir, starting only from basic available data, with a limitation of a reliable PVT data base.