Research On Data-driven Fault Diagnosis Methods Of Polymerizer Batch Process

PVC production is a large scale system which has complex structure, manyfactors can cause the system work abnormally, and each factor is often nonlinear andcoupling relationship. If one of the production conditions abnormal, it can make thesystem collapse or serious impact on the quality of the products, so the fault diagnosisof polymerization kettle is the key to determine the quality of PVC products.Therefore, it has very important meaning for polymerization kettle to fault diagnosis;this text proposed the method that combining artificial neural network withoptimization algorithms to diagnose it. In this text, the main works as follows:There understand the production process of PVC and the basic principle of faultdiagnosis and application of neural network, it detailed introduces the basic structureof SOM neural network, learning algorithm and particle swarm optimization (PSO)algorithm, the SOM neural network is optimized by using the improved PSOalgorithm, the gist of PSO algorithm optimize SOM neural network is that optimizingadjusting weight of all nodes in the survival neighborhood of SOM. By comparedsimulation results, the effect of improved PSO algorithm optimize SOM neuralnetwork is better.There introduces probabilistic neural network (PNN) and the PNN neuralnetwork structure, learning algorithm and rough set theory, in the training process, thefirst reduction on the original data is method by rough set, form a simple set of rules,it make the probabilistic neural network have good topology structure, the scale of thenetwork is more reduced, learning speed is greatly improved. Finally, to optimize thesmoothing factor by using genetic algorithms, it improves the classification ability ofnetwork.There introduces the support vector machine (SVM) and cuckoo algorithm (CS),Due to the selection of SVM parameters have great influence on results of faultdiagnosis, so there use the cuckoo algorithm to optimize the penalty factor and kernelfunction parameters of support vector machine, and using an improved cuckooalgorithm to optimize SVM for fault diagnosis and simulation research ofpolymerization kettle.In conclusion, the results of simulation show that the fault diagnosis of threekinds of neural network can obtain good diagnosis results and improve thepolymerization of PVC polymerization kettle fault diagnosis and prediction rate; itcan satisfy the requirement of real-time control of polymerization process.