Research On Mode Identification Method For Multi-mode Batch Process Based On Bayesian Statistical Analysis

Multi-mode batch process,as an important production mode in industrial production,has the characteristics of small batch,multi variety and high added value.It is widely used in the fields of pharmaceutical,fermentation,semiconductor processing and so on.However,because of the diversity of batch process production schemes and the multi-operation features of the process,the process has multimode characteristics.Realizing the identification of the batch process can divide the process into multiple modes with different characteristics,which is the basis for the abnormal detection of the process measurement data.The traditional mode identification method ignores the timing characteristics of data and does not consider the empirical risk of modal identification,resulting in low identification accuracy.The Bayesian statistical analysis method can combine the prior knowledge of the process with the process data,and use the minimum posterior risk criteria to realize deep mining of process data.Therefore,it is of great theoretical significance and application value to study the modal identification method of multimode batch processes.In this paper,on the basis of analyzing the characteristics and data characteristics of batch process,a multi-mode batch process mode identification method based on Bayesian statistical analysis is proposed.Firstly,in the fuzzy C-means clustering algorithm,the comprehensive index of process data division effect is introduced,and the historical data of the batch process is clustered and analyzed to obtain the optimal mode number of the historical data set.Then the transition mode determination thr-eshold was set,and the mode membership rules were proposed to obtain the results of coarse division of the process mode.After that,a mode inference risk function based on timing constraints was introduced and a Bayesian network classifier was constructed.The final attribution of process modality was decided by using the minimum posterior risk criterion,and the identification of the mode for batch process was realized.Finally,the detection of abnormal process measurement data was achieve after using different abnormal detection methods to construct multiple evidence sources.The number of optimal mode under historical data set could be determined by introducing process data division effect comprehensive index.At the same time,the rough division result of process mode could be obtained after setting the transition mode determination threshold and establishing the mode membership rules.Combining the mode information with the timing constraints of the process data,and introducing the modal inference risk function based on timing constraints,a Bayesian network classifier was constructed.After using the minimum posterior risk criteria to make decisions,the mode recognition of batch process was achieved with the high recognition accuracy.On this basis of that,the sources of evidence were achieved by using different abnormal detection methods.After fusing them together and establishing detection rules,the abnormal detection of process data were finally realized.The experimental research shows that the proposed the mode recognition method based on Bayesian statistical analysis can effectively identify the mode of batch process and has higher reliability and stability than existing cluster analysis-based methods.The proposed abnormal detection method based on DS evidence theory could solve the misjudgment problem of process data abnormal detection,and could improve the detection accuracy.