Research On Monitoring Method For Batch Process Based On Neural Network And PCA Model

As an important production mode in the modern industrial production process,batch process has been widely used in many fields.It can be said that the production of batch processes is closely related to the life of human beings,meanwhile,the safe and reliable operation of batch production process has become the focus of attention in the industry.To handle the non-linear,non-Gaussian and uneven length problems in batch process monitoring,a new monitoring method for batch processes based on neural network and PCA algorithm(NNPCA)is proposed in this thesis.The main ideas of the algorithm are as follows:In order to handle the uneven length problems of batch process data,the three-dimensional data matrix is expanded into two-dimensional data by cutting the variable,so that the batch process data can be processed as the continuous process data.For the non-linear problem of batch process data,then the two-dimensional data are mapped to the high-dimensional space by neural network,and meanwhile the non-linear data is linearized in the mapping process.According to the PCA algorithm,the process data are consistent with the Gauss distribution.Because the negative entropy can effectively reflect the level of the normal distribution,the sum of negentropy of each variable of the neural network outputs as the objective function,using genetic algorithm to optimize the weights and thresholds of the neural network,so that the outputs of the neural network data in accordance with the Gauss distribution.Using neural network output data to establish PCA model,the control limits of SPE statistic and Hotelling T2 statistic are determined and they can be used for on-line monitoring.In the aspect of fault tracing,because the neural network is an explicit mapping,this thesis uses the method of SPE contribution plot to determine the fault variables and draw the contribution plot.Multi stage is the inherent characteristic of batch process,but a lot of batch process does not have definite signs of division stage.For the problem of multi stage classification,this paper use the improved weighted fuzzy C means clustering algorithm to classify the data of a batch of normal working conditions,and the optimal weights and initial cluster centers are obtained by genetic algorithm,which is used as the classification model of the batch data.Then,the classification model is used to classify all batches of batch process data,and the NNPCA model is established in each class.When the on-line monitoring is carried out,the distance discriminant analysis of Euclidean distance is used as the selection strategy of online monitoring model.To verify the validity of algorithm,the Simulink simulation model of electro slag remelting process is established in this thesis.The simulation platform is used to generate data,and the NNPCA algorithm is simulated and verified.In addition,the multi-phase NNPCA model is simulated and verified by using the data of penicillin fermentation process.