Research On Modeling And Correcting Methods Of Slow Time-varying Objects Based On Monitoring

The slow time-varying parameters exist extensively in many fields such as process control and fault diagnosing. As the extensiveness and complexity of slow time varying objects, the limitations of access to information from slow time-vary ing objects, modeling slow time-varying objects is always a hotspot. Therefore, research on modeling and correcting slow time-varying objects has important theory significance and practical value.Firstly, the paper summarizes the research status of modeling algorithm of time-varying systems and process monitoring method based on data-driven.Then it presents PLS modeling and PCA monitoring theory in detail. The paper makes a study of the issue of monitoring the model of a slow time-varying object. According to the model characteristics, the deviation between the actual output and the prediction output of the model and its change are chosen as PCA monitoring variables. A method of monitoring and identifying the model based on change of T2statistic is proposed through a large number of simulations. For general nonlinear slow time-varying objects, it is verified that the monitoring algorithm could monitor model exception before the deviation overruns maximum of deviation in training set through simulations. It would provide advantage to correct model accurately. Based on monitoring the model, the model is corrected before the model error reaches its threshold for lagging of conventional method of correcting model and then the correction accuracy is improved. Meanwhile, in order to reduce the time of correction, a method of correcting model based on combining correcting vector and PLS is proposed for slow time-varying objects through extracting information from adjacent PLS models. It reduces frequency of correcting model by substituting matrix multiplication for correcting model by PLS, and also reduces the time of correction with the prerequisite of guaranteeing the accuracy. The effectiveness of the method which is a combination of correcting vector and PLS for general nonlinear slow time-varying objects is verified through simulations.