Robust Tools for Approximating In-Situ Combustion Process Performance

In-situ combustion process is the oldest and high efficient thermal oil recovery method. Despite its immense effect on oil recovery, the process has not found widespread acceptance among operators since the prediction of its performance is still a problem. So a model should be presented to predict the oil recovery factor obtained by this process as accurate as possible. For this purpose, a Multi-Layer Feed-forward neural network and also an Adaptive Neuro-Fuzzy Interference System (ANFIS) are used in this study to simulate the recoveries of 14 sets of oil field data obtained from literature. Twelve sets of data were used for training and two remaining sets were used to test the models. Comparison between these models and other methods like Gates and Ramey method and second Satman correlation indicates that the errors of ANFIS model for training and test data are lower than that of other methods. Also, these models do not require parameters such as fuel concentration, amount of air required for burning the fuel and also oxygen utilization which are necessary in other methods. Finally, in-situ combustion process recoveries for two Iranian oil fields are predicted by the best proposed model, i.e. ANFIS, as two case studies.

Keywords:

In-Situ Combustion Recovery Factor , Neural Networks , Adaptive Neuro-Fuzzy Interference System , Gates and Ramey Method , Second Satman Correlation