Prediction of Crude Oil Viscosity for Pipeline Operation Safety Using Optimized GRNN Model

Abstract In the whole process of crude oil storage and transportation, the obtaining of the viscosity of crude oil timely and accurately is very important for the safe operation of crude oil pipelines. This paper proposes a new method to predict crude oil viscosity based on Generalized Recurrent Neural Network (GRNN), Particle Swarm Optimization (PSO) and K-Means Clustering Algorithm, which are different from the normal measurement in the laboratory. Feature selection, focusing on obtaining most closely related components to the viscosity of crude oil at 50 °C from the content of Kantian residual carbon, sulfur, nitrogen, water, salt, acid value, ash, colloid, asphaltene, wax, Fe, Ni, Cu and V, is introduced by using the Maximal Information Coefficient (MIC) and Sequential Floating Forward Selection (SFFS) Algorithm. The density and pour point, which are easily available, are determined as the final input of the GRNN by physical property correlation analysis to replace the most related components obtained. For most practical engineering problems that only have a small data set with partial poor samples, the K-Means Clustering Algorithm is introduced in the model to analyze the distribution characteristics of the samples and avoid the impact on the prediction results caused by poor data set. It effectively makes model evaluation more convincing. A Generalized Recurrent Neural Network (GRNN) with the parameter optimization by Particle Swarm Optimization (PSO) is constructed as prediction model. The final results show that the optimized GRNN prediction model in this paper has a good performance on prediction accuracy and stability at the viscosity of crude oil.