Abstract
The objective of this paper is to present some novel applications of the artificial neural network technology in the field of petroleum reservoir engineering. In particular, the paper will be centered in an artificial neural network (ANN) approach to the derivation of nonlinear empirical correlations relating field information, from production tests (GOR, API), with molar composition obtained from validated PVT analysis. As it is well known artificial neural networks are computing devices made of many simple highly interconnected processing units which mimic the information processing that takes place in the neural system of animals. The Radial Basis Function (RBF) neural network architectures have been used successfully in the generation of non-linear correlations between input and output data sets. In this work we use the RBF paradigm to implement a prediction system that relates the gas-oil ratio (GOR) and API gravity with corresponding the molar composition C1 and CO2 and the pseudocomponent composition C2–C6 and C7+. The development of such a prediction system is of great importance since it allows the generation, using the state equation, of synthetic PVT analysis.
This work was made with carefully selected data from two Venezuelan regions with marked differences in fluid properties. On one hand, the new production areas of the North of Monagas and on the other, the traditional old production areas. The neural network training was performed with a data set of 36 examples for the North of Monagas region and 80 examples for the traditional areas. The trained system, was validated with 80 PVT laboratory tests from Eastern Venezuela showing an excellent performance in the prediction of molar composition. In general a prediction error of less than ten per cent was observed.
