Investigation of various characterization methods using generalized distribution model and artificial neural network

• A new approach based on Riazi’s generalized distribution model was developed.
• Input data are SG7+, MW7+ and TBP distribution.
• Also, an ANN network developed for prediction of properties distribution of C7+.
• Prediction with input data (MW7+,SG7+,TBP7+) presented the most precise results.

Characterization of complex mixtures is a key factor in the phase behavior study of reservoir fluids. In this work, a statistical analysis has been carried out over wide range of experimental data to investigate the accuracy of various characterization methods based on the attainable input data of heptane plus fraction (C7+) using generalized distribution model. On the other hand, a new accurate approach was developed based on Riazi’s generalized distribution model with input data of molecular weight (M7+), specific gravity (SG7+) and true boiling point distribution (Tb). In addition, an artificial neural network has been trained and tested for three different sets of input data including (M7+,SG7+), (M7+,SG7+, Refractive index) and (M7+,SG7+,Tb7+). The last input data presented the most precise results in a good agreement with experimental data. Ultimately, the impact of characterization methods and lumping on the phase behavior for two crude oil samples was also investigated.