| In oil refining industry,there are many important quality index which is unable or difficult to measure online,such as 95 point of diesel produced by atmospheric and vacuum distillation is one of the important indicators to measure its quality,diesel’s 95 point can’t be obtained real-timely by the existing measurements.Aiming at this phenomenon,this paper took some petrochemical company’s atmospheric and vacuum distillation unit as the research object,obtained the data from industrial field,builded RBF neural network soft measurement model of diesel 95 point,and builded support vector machine(SVM)soft measurement model of diesel 95 point.Providing the basis and conditions for atmospheric and vacuum distillation unit’s advanced control.This paper presented two kinds of soft measurement models of diesel’s 95 points and verified their effectiveness by simulation experiment.Firstly,the paper elaborated the soft measurement’s engineering background,the modeling method of soft measurement technology and development.Secondly,this article discusses the training methods of hidden layer of RBF neural network at the same time,using the subtraction clustering and K-means algorithm can make the model predicts satisfy industrial requirements better,the model prediction error in the range of industrial requirements.In addition,in order to ensure the accuracy of soft measurement,paper used two kinds of optimization algorithm to set the parameters of support vector machine(SVM)model.They are genetic algorithm(GA)and particle swarm algorithm(PSO).Then crossover and mutation process of genetic algorithm is improved,the simulation experimental results show that the obtained parameters can make the support vector machine has good generalization performance. |
