Prediction of multiple complex crude oil condensation points based on least squares support vector machine

Abstract

The prediction of multiple complex crude oil condensation points has attracted great attention in oil and gas industry. In this paper, firstly, a least squares vector machine parameter optimization method based on Bayesian algorithm is proposed, and a prediction model is established and applied to predict the condensation point of various complex crude oils. Then, the proposed algorithm is compared with the existing condensation point prediction methods and the standard Least squares support vector machine (LSSVM) model. Thirdly, it confirms the feasibility of the proposed method for actual production processes, ensuring the safety and stability of crude oil transportation. Results show that: (a) The Bayesian algorithm uses probabilistic surrogate models and collection functions to effectively optimize the parameters of the model to the most suitable values. (b) The LSSVM algorithm based on Bayesian optimization increases the computation time and maximum absolute deviation (MAD) of the samples. (c) Although the method proposed in this article slightly increases computation time, it can achieve higher accuracy.