Hybrid of Two Heuristic Optimizations with LSSVM to Predict Refractive Index as Asphaltene Stability Identifier

Abstract

Asphaltene precipitation is a big challenge in the petroleum industry. This motivated us to develop a reliable model between refractive index and SARA fraction as a tool for the diagnosis of asphaltene stability. Least-square support vector machine (LSSVM), due to its several unique advantages, has been successfully verified as a predicting method in recent years. However, the success of LSSVMs depends on the adequate choice of the kernel and regularization parameters. We proposed the combination of two search algorithms to deal with the problem of support vector machine parameter selection. On this basis, we combined coupled simulated annealing (CSA) and the Nelder and Mead Simplex method to optimize the parameters. In this hybrid optimization, first, CSA finds suitable starting values and these are passed to the simplex method in order to tune the result. The LSSVM results are promising and accurate, and outperform both neural network and empirical models existing in literature.

Keywords:

• Asphaltene stability
• coupled simulated annealing
• least-square support vector machine
• Nedler–Mead simplex method
• refractive index

Notes

Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/ldis.