Artificial intelligence approach to predict drag reduction in crude oil pipelines.

The addition of a small amount of drag reducing agents (DRAs) to a flowing fluid in a pipeline causes reduction in pressure drop through the pipeline. As a result, energy consumption reduces for a given flow rate. Drag reduction percent (DR %) is an important parameter, which indicates the efficiency of DRAs. Since different variables affect DR % in crude oil pipelines in a complex manner, presenting an appropriate model which can predict such behavior is very beneficial, especially in oil industry. In this research, the multi-layer perceptron (MLP) and radial basis function (RBF) models were presented for this case using MATLAB software. The correlation coefficient (R), root means square error (RMSE), and average absolute relative deviation (AARD) were utilized for assessing the proposed models. The AARD of the MLP model was 4.86% and 7.80% in training and testing stages, respectively. The AARD of the RBF model was 1.11% in training stage and 12.49% in testing stage. Finally, the efficiencies of the presented models were studied using new data set in validation stage and were compared with the results of mathematical correlation. In this stage, the MLP model showed an AARD of 7.42%, and the RBF model showed an AARD of 13.96%, which are three to four times less than AARD of the mathematical correlation, stating the high ability of the proposed models in prediction of drag reduction due to DRAs in oil pipelines. Image 1 • Crude oil specifications, DRA con., flow properties and temp. were the inputs. • To achieve optimal MLP model, various parameters of MLP were optimized. • Training function was the most important parameter in MLP model. • MLP and RBF models could predict DR respectively with 7.42% and 13.96% AARD. • AARD of MLP and RBF models were several times less than AARD of mathematical model. [ABSTRACT FROM AUTHOR]