Accurate machine-learning based modeling of apparent viscosity of waxy oils mixed with polymeric wax inhibitors

Abstract

While producing and then transporting waxy oils through pipeline network systems, wax precipitation commonly leads to issues such as wax build-up, gel creation, and decreased flow, negatively affecting pipeline efficiency. To address these problems, the petroleum industry frequently uses polymeric flow improvers to enhance the rheological properties, thereby increasing the mobility and flowability of oils. Experimental methods to measure oil viscosity are both expensive and time-consuming; hence, having reliable data driven frameworks are necessary in this regard. This work presents the use of accurate artificial intelligence methods including K-nearest neighbors, convolutional neural network, multi-layer perceptron artificial neural network, decision tree, support vector machine, adaptive boosting, random forest and ensemble learning to create smart methods for calculating the apparent viscosity of waxy crude oils that are doped with polymeric additives by actual databank sourced from prior studies. A databank containing 526 datapoints was used to both prepare and validate these machine learning algorithms. The predicted results were then evaluated against actual lab data using both graphical and statistical methods. The outcomes showed a strong agreement between the determined and real values for the constructed models. These developed models offer a way of accurate estimation capability bypassing the need for costly, labor-intensive, and time-consuming experimental procedures.