Study Of Nonstationary Process Monitoring Based On Cointegration Theory

Due to the external disturbances and complex structure of process,data from industrial processes often present obvious nonstationary characteristics,so that fault feature is easily covered by the non-stationary trend of variables,which makes it difficult for monitoring abnormal states during real time operation.Therefore,more attention has been paid to the monitoring research of nonstationary process.Cointegration theory is an analytical method for multivariate nonstationary time series in econometrics,which focuses on the long-term equilibrium relationship among nonstationary variables.For continuous nonstationary industrial processes,a nonstationary process monitoring method based on cointegration theory and joint statistical moments is proposed.For nonstationary variables in process,the deviation series of the long-term equilibrium relationship among these variables are obtained by cointegration analysis,which is called equilibrium error,then form a new stationary data set together with other stationary variables.Different from the commonly used statistic and statistic,which only focus on low-order statistical properties of the data,a monitoring statistic containing mean,variance and skewness is proposed to characterize more comprehensive statistical properties.A simulation data set and an industrial process data set are used as case studies to validate the proposed method for nonstationary process monitoring.Compared with both principal component analysis and monitoring methods based on cointegration analysis and statistics,the proposed method provides lower false alarm rate and earlier alarm time.In batch processes,due to the internal mechanism of the process,there is a long-term equilibrium relationship among some nonstationary variables,which means that the deviation of the equilibrium relationship is stationary.Therefore,a batch process can be monitored based on cointegration analysis.When there is a fault in the process,the equilibrium relationship is broken and the statistical characteristics of deviation series changes.With the variance of the deviation series as the monitoring statistic,on-line monitoring of the process can be conducted.The proposed monitoring method for batch process is validated by the benchmark Pen Sim.During the monitoring procedure,batch trajectory synchronization and batch unfolding preprocessing operations are not required in the proposed method,which greatly simplifies the monitoring procedure of the batch process.The study conducted in this thesis shows that the proposed monitoring method based on cointegration analysis is generally applicable to nonstationary industrial processes monitoring and presents good monitoring effect on both batch and continuous processes.