Development of Predictive Model for Wax Formation in Deep-water Pipeline Using Machine learning

• Published in: Journal of physics. Conference series Vol. 2594; no. 1; pp. 12042 - 12051
• Main Authors: Sultana, Farhana, Abdallah, Elhassan Mostafa, Dong, Shaohua, Mukhtar, Yasir M. F.
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.10.2023
• Subjects: Concentration gradient; Deep sea; Deposition; Gas pipelines; Machine learning; Neural networks; Outliers (statistics); Petroleum pipelines; Physics; Pipes; Prediction models; Radial basis function; Wall shear stresses; Water pipelines; Waxes
• ISSN: 1742-6588; 1742-6596
• DOI: 10.1088/1742-6596/2594/1/012042

Abstract Wax deposition in deep sea pipelines can affect oil and gas production. It happens when the pipeline temperature drops below the wax appearance temperature, causing the wax to solidify and stick to the pipeline walls. This paper discusses the limitations of current models in predicting wax deposition in pipelines and proposes the development of more complex, intelligent models that are accurate, resilient, and adaptable to novel input-output scenarios. It also highlights various methods for preventing wax deposition, including the use of chemical inhibitors and thermal remediation techniques. The paper aims to develop a neural network modeling approach for predicting wax deposition potential in producing reservoirs to prevent accumulation before removal. The parameters that were used are wax molecule concentration gradient at pipe wall, temperature gradient at pipe wall, reported wax deposition rate, the pipe wall shear stress, and crude oil viscosity. The radial basis function neural network (RBFNN), is based on the machine learning technique. RBFNN is a good choice for building a predictive wax deposition rate model because it is highly accurate, robust to noise and outliers, computationally efficient, can model complex nonlinear relationships, suitable for real-time applications and requires less training data. The RBF neural network is constructed using MATLAB. The outcomes and experimental data are consistent with a 1.5 percent relative inaccuracy when compared to prior studies. The R^2 value was 0.9906, R value was 0.9953 and the RMSE value was 0.8598. The RBF neural network-based approach for predicting wax deposition rate was successful.