Artificial Neural Network Based Model for Crude Oil Viscosity Prediction

Viscosity is a crucial physical property of crude oil which is used in the calculations of formation evaluation, fluid flow through porous media and the design of production and surface facilities, and pipeline. A feed-forward back-propagation neural network model with with Levenberg- Marquardt training algorithm is presented based on 357 data sets of Iranian crudes for estimation of saturated and undersaturated oil viscosity. The developed model is tested and compared to some empirical corrolations by 90 data sets. The neural network model is generally more accurate than correlations. It outperformed corrolations with highest corrolation coeficients and lowest average absolute relative errors.