Study On Fault Diagnosis Method Of Chemical Process Based On Non-homogeneous Dynamic Bayesian Network

In industrial processes,data-driven methods are commonly applied for fault diagnosis and detection.Among them,Bayesian networks have received widespread attention because this approach can model complex chemical processes based on data,learn a directed network,and provide intuitive results.With the modeled network structure,fault nodes can be identified and the fault propagation path can be traced.Unlike traditional Bayesian network modeling methods,this is proposes a fault diagnosis method for chemical processes based on a non-homogeneous dynamic Bayesian network(NH-DBN)that focuses on the following:1.The traditional NH-DBN model was studied,and it was found that the model had randomness in the parent node sampling stage,and the selection of the candidate parent node set did not consider the correlation between nodes.This leads to a decrease in sampling efficiency and,to some extent,a lower network construction accuracy.To address this problem,this is proposes a NH-DBN model based on parent node set filtering(PF-NH-DBN),which mainly improves the parent node inference stage.This method uses mutual information and time-series conditional mutual information for parent node filtering,reducing the search space of the initial candidate parent node set and improving the network construction accuracy.Experimental results demonstrate that the PF-NH-DBN model has significantly improved network construction accuracy compared to the traditional model while ensuring model stability.2.The research on the fault diagnosis analysis method based on the NH-DBN model was carried out.The chemical process has many variable nodes and complex network structure.When mutual information is used to screen the parent nodes,in order to avoid the model falling into the local optimal state,the Gaussian mutation strategy based on mutual information is used to further improve the PF-NH-DBN model,effectively improving the global optimization capability.At the same time,a chemical process fault diagnosis scheme based on the improved PF-NH-DBN was proposed.First,the data set for the chemical process was preprocessed and standardized.Second,the chemical process variable relationship network was constructed based on the improved PF-NH-DBN model.Then,the fault nodes were preliminarily identified based on fault analysis indicators,and the fault propagation path was analyzed based on the marginal probability of the network,which provides a basis for fault isolation.Finally,the TE process dataset was used for experimental verification.The results show that the proposed scheme can infer the fault nodes and fault paths and is a feasible and effective method that can provide strong support for industrial process fault diagnosis.