Search

BPNN structure to predict the Apparent Viscosity of Waxy Crude Oil The Back Propagation Neural Network (BPNN) is optimized with Genetic Algorithm (GA). All the chromosomes are decoded to calculate their respective fitness. The high fitness chromosomes will be selected and a better population will be obtained through genetic algorithms. After meeting the requirement (maximum number of generations or goal error), the training will stop and the optimal chromosome can be obtained. The six parameters (K 0 , n 0 , K j , n j , s g and c spw) are input variables of GA-BPNN model. The consistency coefficient (K T) and the flow behavior index (n T) are the output of the model, by which the apparent viscosity of PPD-treated crude oil with shear effect of viscous flow can be determined. • The GA-BPNN model suits different kinds of crude oil, avoids the longtime of experiments and maintains high accuracy. • The entropy generation caused by viscous flow rates is the most important variable in apparent viscosity prediction. • The apparent viscosity should be used with the effect of shear history to design and operate the waxy crude oil pipeline. This study investigated the apparent viscosity of waxy crude oil treated with pour point depressant (PPD). It considered the shear history and thermal history as the main factors to affect the apparent viscosity of the crude oil when transported by a long-distance pipeline. According to previous practice and present laboratory works, the apparent viscosity that can be determined according to the conventional test specifications without taking into account the effect of shear history cannot be used to successfully design and operate a waxy crude oil pipeline. Thus, with the help of entropy generation (s g) combination, which is caused by the viscous flow of crude in the pipeline and the back propagation artificial neural networks (ANN) optimized by a genetic algorithm, a prediction model was developed to determine the viscosity of PPD treated waxy crude oil, which was affected by shear. The performance of the model was evaluated through four statistical indices, such as the Mean Absolute Percentage Error (MAPE). The MAPE of all data of the apparent viscosity was 12.20%. The influence of each variable on the apparent viscosity was investigated through a sensitivity analysis, which revealed that s g caused by viscous flow rates was the most important variable in viscosity prediction. [ABSTRACT FROM AUTHOR]