Research On Fault Diagnosis Based On Nonlinear Dimensionality Reduction Method And SOM Algorithm

The modern chemical process is becoming more and more complicated,and the fault diagnosis in the process is more and more concerned by people.How to carry out effective fault diagnosis and monitoring is still a long-term and a long-term subject.The Selforganizing Maps(SOM)algorithm can map the input data to the plane,so that the same kind of data is aggregated,different types of data are separated,and the topology of the data set is kept unchanged,so that fault diagnosis can be realized.Claim.The increasing complexity of the process industry has led to a large amount of data and a tighter relationship between variables.A single SOM algorithm has been difficult to meet the needs of fault diagnosis.Aiming at this problem,this paper proposes two improved schemes combining nonlinear data reduction algorithm and self-organizing feature mapping neural network to improve the accuracy of fault diagnosis and improve the visualization effect of diagnosis.The specific research is as follows:(1)Aiming at the problem that high-dimensional nonlinear data features are difficult to identify and effectively classify,a fault diagnosis algorithm combining Kernel Principal Component Analysis(KPCA)and SOM is proposed.The algorithm maps the original data to the high-dimensional space through the kernel function,and then performs the principal component analysis method for dimensionality reduction in the high-dimensional space.When facing the nonlinear high-dimensional data,the step of the nuclear mapping effectively improves the non-data.Linear features,which better reflect the characteristic information of different faults,thus improving the fault mapping results of a single SOM to some extent.(2)Considering the relationship between data categories,Kernel Fisher Discriminant Analysis(KFDA)is proposed in combination with SOM fault diagnosis algorithm.After the data is nucleated to improve the nonlinear characteristics of the data,Fisher discriminant analysis is introduced to reduce the dimensionality of the data,so that the data projected into the new subspace has a large inter-class distance and a small intra-class distance..Simulation studies show that KFDA-SOM has a good fault clustering effect in the face of small variance variation data that other algorithms are difficult to handle.(3)The above two fault diagnosis methods are applied to the Tennessee Eastman Process(TEP).Simulation studies show that these two methods can improve the visual effect of fault diagnosis of the original method and improve the correct fault diagnosis rate.Can be applied to multiple types of faults to monitor complex chemical processes.