Research On Soft Measurement Method Of Diesel Fuel Quality Index Based On Random Configuration Neural Network

For oil refining enterprises,product quality is the most important index,which determines the income and fate of the enterprise.However,in the actual production,the product quality is mostly obtained through sampling,laboratory analysis,which can not guide the production operation in time.For example,the quality indexes of oil products,such as dry point,flash point,initial distillation point,etc.,can not be measured on-line and real-time through the existing measurement methods.In order to solve the closed-loop control of product quality,soft sensor technology has developed rapidly,and it has been applied and developed in industry for many years.Researchers at home and abroad mainly focus on modeling algorithms in the field of soft sensor.With the development of modeling technology,soft sensor technology is more and more widely used in industry.In this paper,the on-line soft sensor technology is studied for the important quality index of 95%diesel oil produced by the third line of atmospheric tower in the process of atmospheric and vacuum distillation.It is studied from the aspects of process principle,determination of auxiliary variables,data preprocessing,modeling method,model correction,soft sensor instrument implementation and so on.Starting from the application background and difficulties of soft sensor implementation,this paper takes the atmospheric and vacuum distillation of the first processing process in oil refining production as the research object;Through the experimental test,the field data processing and mining are carried out to determine the auxiliary variables;the traditional modeling method based on regression analysis is analyzed;finally,a radial basis function(Radial-Based Function,RBF)neural network model is established.In view of the fact that the model is sensitive to the network structure,biogeography optimization algorithm is used to optimize it.However,it is still found that the model relies heavily on prior knowledge and is difficult to implement in the field.Therefore,this paper establishes a diesel quality index soft sensor model based on random configuration neural network,which has accurate learning ability and excellent pan-Chinese ability.The most important feature of the neural network is that the structure of the neural network does not need human intervention.Under the constraint of the general asymptotic approximation inequality,the number of nodes is self-increasing,and the bias and weights are randomly assigned.Through the realization test of the soft sensor system,the application effect is good in the field;the application results show that the 95%diesel soft sensor model based on random configuration neural network has good generalization performance.It provides the basis and conditions for the closed-loop control of diesel oil quality in atmospheric and vacuum unit.