On the determination of crude oil salt content: application of robust modeling approaches

Existing salt content in crude oils can lead to serious problems during petroleum production, refining process, transportation and also some related chemical and petroleum engineering processes. Therefore, a reliable and accurate model is required to predict crude oil salt content for overcoming the problems faced. In this work, an accurate predictive model is presented on the basis of soft-computing methodology, namely the least square support vector machine (LSSVM) for the calculation of crude oil salt content of about 63 samples gathered from the literature. Additionally, a previously published correlation as well as a multilayer perceptron artificial neural network (MLP-ANN) is used for comparing and validating the results obtained by the LSSVM model presented in this study. Moreover, to detect the outlier and/or suspected data point (s) available in the dataset, a reliable approach (Leverage strategy) is implemented in the present study. Our conclusions show good agreement between the actual and values of salt content predicted by the LSSVM model in comparison with the studied methods. Consequently, the R-squared error is 0.9999 and also an overall average absolute percent relative error equal to 0.0093 is reported for the LSSVM model developed in this study.