Investigating a machine learning algorithm's applicability for simulating the apparent viscosity of waxy crude oil in a pipeline

Accurately estimating the formation of viscosity is a vital aspect of pipeline functioning. A study was undertaken here to investigate the precision of utilising a machine learning system for predicting the viscosity of waxy oil in a pipeline environment. A neural network model was developed to ascertain the viscosity of waxy crude oil based on a collection of eight independent parameters. The network model, derived from 30 experimental data points, consists of a hidden layer including 14 neurons. An accuracy analysis was conducted by comparing the predicted viscosity of the network model to the experimental viscosity. The model was built via the Levenberg-Marquardt training algorithm. The accuracy of the artificial neural network's predictions was assessed by calculating the mean squared error value of 2.75 × 10 –3 and the correlation coefficient of 0.99850. The model's anticipated viscosity values had an average deviation of 0.5%. The experiment yielded conclusive evidence that the specifically engineered artificial neural network successfully forecasted the viscosity of waxy crude oil within the pipeline with great precision. [Received: November 15, 2023; Accepted: May 9, 2024]