Research On Fault Diagnosis Method Of Distillation Column Based On Convolutional Neural Network

In the process of petrochemical production,distillation column is an important heat and mass transfer equipment,and its operation directly affects the economic operation benefits of refinery enterprises.Distillation column is the key equipment in the production process of refining and chemical industry,and it is also the hotspot of current research.Research on its fault diagnosis is not only helpful to improve production efficiency and economic benefits,but also valuable to theoretical research.In this paper,the azeotropic distillation column for acetic acid dehydration was studied,and its process was simulated by Aspen Plus software platform.In view of the non-linear,time-varying and interrelated characteristics of chemical process variables,this paper uses the kernel principal component analysis(KPCA)algorithm to extract features from the collected data and determine whether there is a fault.A fault diagnosis classification model of support vector machine based on improved genetic algorithm(IGA)and a fault diagnosis classification model based on convolutional neural network(CNN)are established to complete the acetic acid treatment.Fault diagnosis of dehydration azeotropic distillation column.The main contents of this paper are as follows:(1)The theoretical principle and algorithm process of the nuclear principal component analysis(KPCA)algorithm are studied,and the method of applying KPCA method to non-linear process monitoring is analyzed.The experimental simulation takes the acetic acid dehydration azeotropic distillation process as the object and monitors it with KPCA method.The simulation results show that KPCA can monitor the acetic acid dehydration azeotropic distillation process.The monitoring effect is better.(2)Aiming at the difficulty of parameter selection of support vector machine,an improved genetic algorithm(IGA)is proposed to optimize its parameters.IGA uses generation gap selection and crossover probability to ensure that the most suitableindividuals in the current population are always propagated continuously to the next generation,and make the object of optimization in the later stage of evolution easier and more stable,and improve the computational efficiency.Based on improved genetic algorithm,a multivariable process fault diagnosis method based on KPCA-IGA-SVM is proposed in this paper.The simulation study on the fault condition of acetic acid dehydration azeotropic distillation process shows that the proposed algorithm is effective.(3)In view of the shortcomings of shallow machine learning algorithm in fault diagnosis,a fault diagnosis classification model of distillation column based on convolution neural network(CNN)is established in this paper.In order to further evaluate the model constructed by CNN and the influence of training mechanism and structure results on it,this paper constructs the fault diagnosis classification model of distillation column based on support vector machine(SVM),artificial neural network(BP)and depth confidence network(DBN),and draws the conclusion that the fault diagnosis classification model of distillation column based on CNN is effective through the comparative study of experimental simulation.It is better and has good practicability.