Research On The Viscosity Related Properties Database Of Crude Oil And The Prediction Of Crude Oil Viscosity

The viscosity of crude oil is a very important parameter to determine whether the oil from the wharf can be unloaded.When the offshore oil tanker unloads the oil from the wharf,if the viscosity of crude oil is too high,it will bring significant security risks to the unloading,storage and subsequent pipeline transportation of the crude oil.Therefore,it is of great practical significance for the safe operation of crude oil pipelines to actively explore the correlation between crude oil components and the influencing factors of crude oil viscosity-temperature relationship and to propose a model for the prediction of the viscosity-temperature relationship of crude oil with high accuracy,good universality and simple operation.At present,the research on crude oil viscosity prediction model at home and abroad mainly focuses on fitting the experimental data of crude oil in the state of newtonian fluid,and the viscosity calculation fitting formulas proposed are mainly for the field of specific crude oils.There are few researches related to the establishment of the model for the prediction of viscosity-temperature relationship of crude oil based on machine learning algorithms by using field data,and it is limited to the prediction of viscosity at a certain temperature of crude oil in newtonian fluid.From this perspective,the purpose of this paper is to predict the total viscositytemperature relationship of crude oil by using the clustering algorithm and the optimized neural network method with the imported crude oil from Sinopec wharf as the research object.The viscosity related properties database for the prediction of crude oil viscosity is developed to save and accumulate relevant data,providing a platform for the increase of subsequent data volume.According to the influence between crude oil components and crude oil physical properties,the content of residual carbon,resin,asphaltene,sulphur content,wax content and Ni as the characteristic combinations of crude oil components with the closest correlation with crude oil viscosity in the newtonian state is obtained by the maximal information coefficient(MIC)and sequential floating forward selection(SFFS).And according to the actual situation of field prediction and physical property correlation analysis,the value of density and pour point are used as new characteristics to replace the above characteristics,and which is verified.In view of the rheological characteristics of crude oil at low temperature,the infrared spectrum data and differential scanning calorimetry data of crude oil are introduced.One-dimensional convolutional neural network and piecewise integral are used to obtain feature vectors respectively,as supplementary feature vectors for viscosity prediction of crude oil in nonnewtonian state.And the results show that the viscosity-temperature coefficient of crude oil can be accurately predicted by adding the infrared spectrum and cumulative wax precipitation as the input vector of the prediction model,and then the viscosity-temperature relationship of crude oil is obtained,and the accuracy is much higher than that only using the value of density and pour point as the feature vector,which verifies that the viscosity of crude oil in nonnewtonian state and newtonian state needs to be predicted by different feature vectors.In view of the small amount of data and uneven data distribution at the beginning of the creation of the crude oil viscosity-related physical property database,the distribution characteristics of the data set is analyzed by the K-Means clustering algorithm to explore the influence of data distribution on the prediction accuracy and reduce the influence of the small size and uneven distribution of data set on the prediction model.Based on the rheological correlation formula of crude oil,the viscosity-temperature coefficient used to characterize the viscosity-temperature relationship of crude oil is regressed by using the viscosity-temperature data of crude oil,which provides the possibility for the overall prediction of the viscositytemperature relationship of crude oil.The characteristics of crude oil in both newtonian state and non-newtonian state are used as input feature vectors,respectively.The GRNN is optimized by QPSO algorithm,and the viscosity-temperature coefficients of crude oil are predicted by QPSO-GRNN algorithm,so as to obtain the apparent viscosity at any shear rate and temperature.And the prediction results show that the QPSO-GRNN algorithm can achieve good prediction results.Based on the constructed crude oil viscosity prediction model,a simple graphical client operation interface is developed to facilitate the prediction operation of crude oil viscosity,and it is integrated into the crude oil viscosity-related physical property database as a prediction module.