As one of the most important industrial production,the batch process has been theoretically researched and applied to industry,especially produced high value products in some industries,such as chemical,pharmaceutical,biological,semiconductor and so on.In order to ensure the safe and stable operation of the batch process,it is necessary to monitor the disturbances and the artificial misoperation in the system.Because of the special properties of the batch process,the difficulty of the process monitoring is increased,such as dynamic characteristics,nonlinearity and multi operation stages.Therefore,it is of great theoretical significance and practical value to study the monitoring method of batch process.In view of the common characteristics of batch process,this study focuses on model-based monitoring methods.Firstly,a hybrid model based on the combination of mechanism model and support vector machine(SVM)is studied.Based on the structural risk minimization,the model of batch process is modeled.Secondly,based on the results of the mixed model,the cubature Kalman filter(CKF)is used to preprocess the data.The state new interest sequence and the observation and prediction residual sequence are obtained.Finally,the data sequence is processed by the canonical variable analysis(CVA),and the interval fault monitoring is completed according to the monitoring control limit.The proposed monitoring method was applied to the typical batch process fermentation process,and the expected results were obtained.The hybrid structure model based on SVM and mechanism proposed by this project has good robustness.The CVA method is used to monitor the CKF filtering innovation sequence,which can effectively reduce the leakage rate of the system fault. |