Research On Application Of VAR-PCA Method In Industrial Process Monitoring

Process monitoring plays an increasingly important role in modern industry.In the actual industrial production process,process monitoring can realize early identification of abnormal working conditions,thus preventing accidents and reducing economic losses of enterprises.The methods involved in process monitoring research can be divided into three categories:mathematical model-based,knowledge-based and data-driven methods.Although the monitoring method based on process mechanism or qualitative knowledge has clear physical significance,with the increasing complexity of modern industry,it is not easy to obtain perfect process mechanism,and it is difficult for the method of qualitative knowledge include enough possible faults.Therefore,the model established by these two methods may have great deviation from the real situation,and it is unable to get ideal monitoring effect.The data-driven method can effectively avoid the above problems.In recent years,with the continuous development of computer technology,process monitoring models based on data-driven are increasingly favored by industry.With the development of instrument technology and the wide application of DCS system in industry,a large number of production process data are collected and stored,and these data contain rich process information.Through the study of the data,the industrial production operation status can be well grasped,which also provides a good foundation for the data-driven monitoring method.Under normal operating conditions,the relationship between variables in the production process should be relatively stable,and the current data of variables are influenced by their own and other historical data.We can use the data of these variables at different time points to approximate the relationship between variables by multivariate statistical regression method,and the residual error between actual data and fitted data usually accords with the normal distribution of zero mean.When the process is abnormal,the correlation obtained based on the normal operating conditions will change,leading to the change in the deviation distribution of the actual data and the calculated value of the model.Based on this,this paper proposes a process monitoring method based on vector autoregressive-principal component analysis(VAR-PCA),considering the autocorrelation of variables and the cross-correlation between variables in industrial data.VAR can realize short-term prediction of multivariate time series variables under normal working conditions in production process.Firstly,VAR model is used to extract the autocorrelation and cross-correlation information of process variables under normal operating conditions,and then PCA method is used to further extract the main change information of residual error between actual data and model fitting data,and establish statistics.When abnormal operating conditions occur,the residual error changes caused by faults can be amplified,thus realizing early identification of abnormal operating conditions.The process monitoring model is applied to the feed heat exchanger and catalyst regenerator of a catalytic reforming unit,which can achieve early identification effect for abnormal drop of pressure drop of the feed heat exchanger,and can achieve early warning effect of 26 minutes in advance by comparing operation records for regenerator flying temperature failure,thus realizing process monitoring of actual industrial data.