Prediction of Crude Oil Pyrolysis Process using Radial Basis Function Networks

Prediction of Crude Oil Pyrolysis Process using Radial Basis Function Networks

Knowledge of crude oil Pyrolysis and combustion is one of the most important in oil production using in situ combustion method as a section of thermal enhanced oil recovery (EOR) methods. In this method, crude oil undergoes a series of physical and chemical changes that can refer to pyrolysis as a most important part of these changes. In this work, we have developed Radial Basis Function networks (RBFN) models to predict remaining weight of crude oil during crude oil pyrolysis process. API density, viscosity, resin and asphaltene and other components of crude oil content, temperature and heating rate are selected as RBFN input parameters, whereas remaining weight of crude oil in different temperatures is considered as network output. The data were obtained by doing thermogravimetric analysis and separation experiments on six samples of various Iranian crude oils. The results of this work show that using a RBFN, we can predict the remaining weight of crude oil during its pyrolysis process with an average absolute relative error (ARE) 5.88 percent and mean square error (MSE) 6.15 by newrbe function and an average absolute relative error (ARE) 7.25 percent and mean square error (MSE) 2.51 by newrb function for test data. More over, the results of regression analysis showed a very good coincidence between the laboratory results and predicted results by the proposed RBFN.

Knowledge of crude oil Pyrolysis and combustion is one of the most important in oil production using in situ combustion method as a section of thermal enhanced oil recovery (EOR) methods. In this method, crude oil undergoes a series of physical and chemical changes that can refer to pyrolysis as a most important part of these changes. In this work, we have developed Radial Basis Function networks (RBFN) models to predict remaining weight of crude oil during crude oil pyrolysis process. API density, viscosity, resin and asphaltene and other components of crude oil content, temperature and heating rate are selected as RBFN input parameters, whereas remaining weight of crude oil in different temperatures is considered as network output. The data were obtained by doing thermogravimetric analysis and separation experiments on six samples of various Iranian crude oils. The results of this work show that using a RBFN, we can predict the remaining weight of crude oil during its pyrolysis process with an average absolute relative error (ARE) 5.88 percent and mean square error (MSE) 6.15 by newrbe function and an average absolute relative error (ARE) 7.25 percent and mean square error (MSE) 2.51 by newrb function for test data. More over, the results of regression analysis showed a very good coincidence between the laboratory results and predicted results by the proposed RBFN.