Prediction of Total Acid Number in Crude Oil Based on Improved Multi-Kernel Gaussian Process Regression

Total acid number affects the corrosion of crude oil in the process of transportation and processing and is an important index to evaluate the crude oil. In this paper, we propose a prediction model based on improved multi-kernel Gaussian process regression (IMK-GPR). The inputs of the prediction model are nuclear magnetic resonance hydrogen spectrum data and the output is the total acid number of crude oil. First, we use GPR to establish the model between the nuclear magnetic resonance hydrogen spectrum data and the total acid number in crude oil, and the multi-kernel function is used as the covariance kernel function of GPR to improve the adaptability of the model for the data. Then, an improved backtracking search optimization algorithm is proposed to optimize the hyperparameters of IMK-GPR, which can improve the accuracy of prediction model. Finally, the experiments using practical data collected from a refinery are carried out. The results indicate that the proposed model outperforms the other state-of-the-art data-based methods in terms of accuracy.