Fault Identification And Diagnosis Of TE Process Based On Fisher Discriminant And ICS-RBF

Fault diagnosis of chemical industry processes has always been an important part of the entire industry,and it is a strong guarantee for the economic and security of enterprises and countries.At the same time,because the production mode of the chemical industry is doomed to be unintuitive,the state of the production process can only be known through data,and the chemical production process is an environment with large variables,traditionally monitored and troubleshooting method is not ideal.The need for troubleshooting for high-dimensional big data is growing.A high-dimensional big data real chemical fault model based on TE process for the requirements of modern industrial production environment is studied in this paper,and an improved Fisher fault identification method is designed.The ICS-RBF neural network is designed for fault diagnosis based on the stripped fault data.The simulation study of fault diagnosis was carried out.The main work of the thesis is as follows:(1)Referring to the relevant literature and data,the TE process simulation model was established,the fault model output was obtained,15 faults of the TE process simulation model were analyzed in detail,and the evaluation index of the chemical fault diagnosis model was determined.(2)According to the high-dimensional fault of TE process,the fault identification is carried out.Based on the discrete relationship between sample classes and sample classes,the improved Fisher algorithm with penalty factor is designed and simulated.The simulation results show that the proposed method improves the projection coefficient and reduces the weight of the abnormal data in the production process.(3)Fault diagnosis based on ICS-RBF neural network is performed for fault data obtained by fault discrimination.Aiming at the slow training speed in the RBF network,the cuckoo model based on Levi's flight is used to optimize the RBF network parameters,and the cuckoo model(ICS)based on the individual historical optimal solution is designed.The optimal nesting and individual gaps are used to calculate the additional step size,so that the difference between the nesting solution and the global optimal nesting solution are taken into account in the stepping of each generation of nest solutions,which improves the training speed and verifies the effect with the test function.The simulation experiments of a total fault diagnosis and alocal random fault diagnosis is carried out based on the RBF neural network.The simulation results show that the diagnostic effect is good.Finally,the advantages and disadvantages of TE process fault model and fault diagnosis algorithm are analyzed,and suggestions for further research work are proposed.