•  
  •  
 

Keywords

Wax appearance temperature, crude oil characterisation, data analytics, gamma distribution function, machine learning, wax precipitation

Document Type

Research Article

Abstract

The precipitation of wax in crude oil pipelines poses a significant operational challenge for the petroleum industry. This study proposes a novel approach to characterising crude oil and predicting wax appearance temperature (WAT) using advanced data analytics techniques, including group method of data handling (GMDH), support verctor machines (SVM), extreme gradient boosting (XGBoost) and various regression models, as alternatives to conventional artificial neural networks (ANN). Crude oil was classified into four categories: heavy oil/biodegraded, paraffinic oil, waxy oil, and light oil/paraffinic condensate. A detailed characterisation framework based on the gamma distribution function (GDF) was employed to split lumped hydrocarbon fractions for more accurate predictions. Automation of variance calculation within the GDF model ensures faster and more precise determination of GDF parameters alpha (α), beta (β), and variance (η). This research demonstrates that density and wax content are the most influential parameters among the six laboratory data parameters used for WAT prediction, with the gradient boosting regressor (GBR) achieving an R R-squared value of 0.9048 and an root mean square error (RSME) of 3.42. The findings offer a significant step forward in optimising wax management strategies

html

Publication Date

30-9-2025

First Page

20

Last Page

27

Download

Share

COinS