Robust Stochastic Configuration Network Based Prediction Model of Total Nitrogen Content in Crude Oil

The refineries usually process the blending crude oil at present. Blending crude oil leads to the increased uncertainties of physical properties of crude oil. This paper presents a robust stochastic configuration network (SCN) based prediction model for total nitrogen content in crude oil. First, the robust SCN based on Bayesian framework is proposed. In the Bayesian framework, the noise and outliers are assumed to be the mixture of two Gaussian distributions to improve the robustness of SCN. Then, to solve the problem that there are no analytical solutions of parameters due to the assumption of mixture of Gaussian distributions, parameters of the proposed robust SCN are estimated by the expectation maximization (EM) algorithm. Experiments using the real-world data collected from a refinery are carried out. The results indicate that the proposed approach has better accuracy and robustness.