Research On Multi-level Data Reconciliation In Process Industry

Data reconciliation is a traditional technology.After more than half a century of development,the subject has been developed into a cross-disciplinary research field combining statistics,operations research,system modeling,optimization algorithm,management and other fields.With the continuous improvement of industrial automation technology and the rapid development of IT technology,data reconciliation technology has also integrated many new technologies of artificial intelligence.With the expansion of enterprise scale and the improvement of application demand,data reconciliation is facing more severe challenges.It receives higher requirements from process simulation,statistical analysis,fault diagnosis and other applications.Accurate data provided by data reconciliation and significant error detection are the basis for many important upper levels to operate effectively.In the face of new requirements,traditional algorithms need to be improved,and many new problems need to be solved.After reviewing the importance of data reconciliation technology at home and abroad and the overview of theoretical research,this paper makes a thorough study on hierarchical modeling,large system decomposition,multi?level reconciliation,discrete event tracking and dynamic data reconciliation algorithms.The main contents and innovations of this paper are as follows:1.A multi-level date reconciliation framework for hybrid systems is proposed.Based on the analysis of multi-level and hybrid characteristics of process industry data,a hierarchical material balance modeling method with discrete events at different time scales is proposed.The relationship between different levels of data is fully explored,and the production process of modern process industry enterprises is more accurately described Based on the modeling method and the problems in actual production,a multi-level data reconciliation framework for hybrid systems is proposed,which includes single-level decomposition of large-scale systems and dynamic data coordination methods for subsystems.Through discrete event tracking and detection,reliable event information is provided for hybrid systems.Under this framework,the data reconciliation of the whole plant in the process industry can be accurately completed.2.Based on the multi-level material balance model,the difficulty of solving complex systems is analyzed from the point of view of single-level global solution,and the significance of decomposition of large-scale systems is given.Based on the basic principle of graph theory,virtual instrument is used to simplify complex network,cut set search is used to find the best decomposition method,which can reduce the coupling degree between subsystems to the greatest extent,and then simplify the coordination process between subsystems.A large-scale system decomposition method based on graph theory is proposed.The application flow of single-level solution method in multi-level system is also given.3.In order to provide discrete event information for hybrid systems more timely and effectively,a scheduling event tracking and restoring method based on production data is proposed.Based on the analysis and processing of production data,using the redundancy of information between multi-source measurement data and the methods of information integration,event tracking and mobile synthesis,the on-line tracking and restoring of on-site scheduling events in process industry are realized.4.A new robust least squares dynamic data reconciliation algorithm is proposed for the first time.The robustness of Huber estimation is further improved by introducing local redundancy.At the same time,by adding elimination interval to Huber estimation weight function,the most significant error suspected measurement value will be deleted in the iteration process,which improves the accuracy of the algorithm.The recognition rate of significant errors in weak redundant variables has been improved.Based on the above improvements and the advantages of online filtering,a new linear robust least squares data reconciliation algorithm is proposed.