인공신경망을 이용한 석유 회수증진기법 선정 연구

Abstract

Abstract

High oil prices and concerns about future oil supply are leading to a renewed emphasis on the unconventional hydrocarbon and the remaining oil in mature fields. These resources could be recovered by EOR, Enhanced Oil Recovery, which is defined as the reservoir treatment process to inject certain substances, usually not included in the reservoir, to improve the oil recovery. These processes have gained more attention as a result of increasing costs of exploring the new oil fields and reducing opportunities of discovering the high reserve and good quality reservoirs.

Since the oil production by EOR is a rather difficult, risky and expensive, the selection of proper EOR method according to the reservoir condition is important to attain a successful and profitable project. The main tool for predicting reservoir performance after EOR application, hence selecting a proper method, is a reservoir simulation which requires extensive information about the reservoir that may not be available or can be unreliable at the initial evaluation stage and also extensive time. Another method is using an expert opinion, but it tends to be biased by operational experience of the expert.

In this study, the Artificial Neural Network model is developed to solve the technical problem in selecting the EOR method. The model is composed of the four layers which consist of one input layer of the seven neurons, one output layer of the five neurons, the first hidden layer of the ten neurons, and the second hidden layer of the eight neurons. The input neurons contain the main reservoir parameters, and the output neurons contain the EOR methods to be evaluated. The tangent-sigmoid function is used as an activation function of the first hidden layer, the log-sigmoid is used in the second hidden layer, and the pure linear function is used in the output layer. The data used in training and testing the networks are extracted from the special report of Worldwide EOR Survey published by the Oil&Gas Journal at 2006. The network is trained by the scaled conjugate gradient algorithm.

After trained successfully, the noise test is performed to examine whether the model overcomes the error that may be included in the data. After that, the model is tested by data which are not used for training to evaluate the model applicability. Finally, the model is applied to the most successful producing EOR projects. The noise test and applicability test show that the ANN model developed in this study can be used to select the most appropriate EOR process based on the basic reservoir properties in a very short and cost effective way. Technical characteristics, limitations in application, and application ranges of each EOR method presented in the previous literatures are also discussed here as a basis for the model development.