Research On Fault Diagnosis Method Of Chemical Process Based On Complex Network Topology

With the large scale and complexity of chemical production,the big data of chemical process monitoring has the characteristics of multi-dimensional,highly coupled and nonlinear.It is difficult to accurately monitor and diagnose a large number of monitoring variables.In order to effectively extract fault features from the big data of chemical process monitoring,a fault diagnosis method of chemical process based on complex network is proposed to identify faults timely and accurately in this paper.Firstly,the parameters such as degree centrality,closeness centrality,betweenness centrality,eigenvector centrality and constraint of complex network are selected as evaluation indexes of node importance.Entropy Weight Method(EWM)is used to determine the weight of each indicator,and an optimization compromise solution method(VlseKriterijumska Optimizacija I Kompromisno Resenje,VIKOR)is applied to rank the importance so that the key nodes in the complex network can be identified to ensure the normal condition of chemical production.Secondly,the complex network is established by using the historical data under normal working condition.Aiming at the problem that the traditional cosine similarity method can ignore the scale between the data,the adjusted cosine similarity method is introduced into the complex network modeling to fully consider the comprehensive influence of the angle and scale between the data vectors on the correlation.The adjusted cosine similarity method is useful to improve the accuracy of the model and the complex network model established can better reflect the real condition of the system.Then,a condition monitoring method based on complex network is proposed.According to the stability of the complex network,the optimal number of samples is determined in the complex network so that the established complex network can describe the real condition of the interaction among the nodes in the system.Based on the topological structure of the complex network,the node anomaly coefficient and its threshold value calculation method are constructed,which are used to extract the real-time operating condition characteristics to realize the quantitative condition monitoring.The root cause of the fault is diagnosed by the topological characteristic parameters of network nodes under abnormal condition.Finally,the Tennessee Eastman(TE)process is used to verify the effectiveness of the proposed method.The results show that the EWM-VIKOR method is used to rank the node importance of TE process,and the important nodes in TE process are identified successfully.The complex network model based on the improved cosine similarity clearly and intuitively reflects the relationship of the monitoring nodes in the system.Moreover,it can quickly and accurately identify abnormal condition and diagnose fault causes in the complex chemical processes to provide effective decision-making basis for plant operators and managers.