Soft computing method for predicting pressure drop reduction in crude oil pipelines based on machine learning methods

Abstract

In the oil industry, the drag-reducing agent has been used to reduce turbulent friction of fluids. The main effort of this study is to examine the feasibility of four novel machine learning models, namely multilayer perceptron, M5Rules, decision table (DT), and trees M5P to estimate the percentage of drag reduction. Then, the mentioned methods are utilized to identify a relationship between the input and output parameters of the crude oil pipeline system. The parameter percentage of drag reduction was taken as the essential output. In contrast, the input parameters selected the flow rate of oil, polymer concentration, kind of polymer, temperature, as well as pipe diameter and roughness. The predicted results obtained by the tools mentioned above were evaluated according to several known statistical indices, namely coefficient of determination (R2), mean absolute error (MAE), root mean squared error (RMSE), relative absolute error (RAE), and root relative squared error (RRSE) as well as novel ranking systems of color intensity rating and total ranking method. The training and testing results of the DT learning method for the R2, MAE, RMSE, RAE, and RRSE were (0.9616, 3.9008, 5.8698, 24.5259%, and 27.4406%) and (0.8964, 6.937, 10.318, 43.3841%, and 45.6581%), respectively. The obtained results, in analyzing the training and testing datasets, proved that DT is the best predictive network to predict the percentage of drag reduction.