Crude Oil Properties Evaluation Model Based On Improved Particle Swarm Optimization Ensemble Random Weights Neural Networks

The process of analyzing the newly mined crude oil properties through experiments is called fast evaluation of crude oil,also called crude oil properties fast evaluation.It plays an important role in the oil refining process,providing real-time data on crude oil properties for crude oil processing.Crude oil properties evaluation is the core content of the fast evaluation of crude oil,the main properties include API,salt content,sulfur content,total carbon,total hydrogen,octane number,acid number,cetane number and so on,as well as the quality yield of different sections of crude oil,distillate properties and residuum properties.According to crude oil properties processing programs and refining performance of a comprehensive evaluation,which provide basic data and processing programs for frequently decompression process and secondary processing equipment.Therefore,crude oil properties model which is rapid and accurate is of great significance to the entire crude oil industry.This article supported by the National Natural Science Foundation project "The theory and implementation technology of global collaborative optimization operation in refinery production process",the research on crude oil properties fast evaluation modeling method in refinery process.The detailed work is list as follows:(1)A description of crude oil properties rapid evaluation.Among them,we analyze the research status of crude oil fast evaluation;the advantages of using nuclear magnetic resonance(NMR)technology of crude oil fast evaluation;the process and the basic principles of crude oil fast evaluation based on nuclear magnetic resonance technology.We confirmed the relationship between input and output of crude oil fast evaluation based on online NMR.(2)We propose an improved online learning ensemble random neural weight networks modeling method.Based on ensemble random neural networks,using the regularized negative correlation learning(RNCL)to combine the individual neural network;it also shows how to use genetic algorithm to optimize the parameters of activation function of the hidden nodes.In addition,we used particle swarm optimization(PSO)algorithm to optimize the optimal hidden layer node number(L)and the optimal random neural network number(M).Finally,the online learning method is used to learn new data and update the model at the same time,which makes the model have better adaptability.(3)we used the improved particle swarm optimization ensemble random neural weight networks model to analysis the three crude oil properties,such as total carbon,carbon residue and API,and Proposed a rapid evaluation of physical properties of crude oil modeling methods and conducted an experimental study.Crude oil properties fast evaluation model was established by using the actual crude oil NMR spectrum of a refinery industry and the indicator data of crude oil properties fast evaluation.The proposed algorithm is compared with PCAERNN,ERNN,Online-DNNE and PLS models,the experimental results show that the proposed method improves the modeling accuracy and validates the effectiveness of the selection and modeling methods.