Research On The Fault Diagnosis Of Polymerizer Based On Rough Set And Neural Network

Polymerizer is the main equipment of polymerization.Its normal work concerns the key of whether the whole process can be carried out successfully.How to apply the intelligent system to the fault diagnosis in traditional industrial production has become a research hotspot.Through the actual polymerization of various process variables in historical data reactor operation monitoring,this paper proposes a genetic algorithm to optimize the application of the method of rough set and neural network.It aims to train the neural network model of reliable data through effective optimal learning and applys neural network to fault the diagnosis of polymerizer.First of all,The neural network is often used to establish the nonlinear data model.It has the advantages of associative memory and strong anti-interference ability.But the traditional BP algorithm has slow convergence.At the same time,the convergence process is often less than the minimum point,showing the local optimum to suspend the training.After analyzing the improved learning algorithm of the common BP neural network,this paper studies the advantages and disadvantages of various neural network improved algorithms through comparative simulation.Levenberg_Marquardt numerical optimization algorithm is used to improve the deficiency of traditional BP training.In this paper,BP neural network is designed for fault diagnosis of 70m~3 PVC polymerizer in Jinxi chemical industry.However,the diagnostic model does not meet the accuracy requirement of fault diagnosis.Secondly,This paper combines rough set and neural network.Rough set theory and attribute reduction are systematically studied.Attribute reduction is NP-hard.When a large number of data set attribute overlap,the positive domain general method often has a large amount of calculation and can not achieve the optimal.This paper uses genetic algorithm to optimize the attributes system problems and improves attribute dependency structure.It choices necessary attributes to nuclear retention which prevents the information redundancy in the relative reduction in the objective necessary to remove the property.Through this way it gets a set of optimal collection of the best attributes and enhances the flexibility and reliability of the attribute reduction,By reducing the attribute of the genetic algorithm of the rough set,the set eliminates the redundancy of the attribute data.This can effectively reduce the over fitting phenomenon of BP network training and improve the accuracy of fault diagnosis.Finally,a qualitative diagnosis scheme of fault tree is designed through the common problems of the polymerizer in Jinxi chemical industry.And then this paper will combine the rough set and neural network applied to the chemical polymerization unit fault diagnosis device.At the same time,in the design of the initial parameters of the BP network,the initial random thresholds and weights will give an uncertain impact on the training of the entire BP network.This paper presents the design optimization of the BP network initial threshold and weight of the genetic algorithm and maximizes the precision of the test sample.It is effective in fault diagnosis training and simulation results through the neural network,which has a good guiding significance for the fault diagnosis of the actual factory production and the theory.