CCR Performance Monitoring And Its Optimization Based On Multivariate Statistics

CCR is a secondary oil processing unit. It is claimed as an indispensable fundamental production element in a modern refinery and/or petrochemical facilities. As a rather complicated process, it is characterized by uncertainties (i.e. unidentified environmental structure and parameters, time-varying, randomness, mutation), non-linearity, correlations among variates and incompletion or lag of information. Stability of variate correlations directly impacts constant operation process and product quality consistency, hence it is impossible to achieve satisfactory univariate monitoring, not to mention optimization, of various variates. Furthermore, when there is big noise around in operation, it would become even harder to pre-alarm any possible failure(s) through univariate monitoring.In All CCR units in China today are stereotyped with technologies of either UOP, USA or IFP, France. Due to patent /technology confidentiality and insufficient app- rehension of characteristics of specific elements like reactor, regenerator, etc. plus the unavailability of publicly disclosed multi-purpose reactor mechanism model for reference, so far upgrading and improvement of CCR's controllability are still relying on out-sources. Taking into consideration that a CCR is of relatively more variates for process monitoring and its DCS is able to generate a large number of real data, this article builds up a data-driven mathematical model based on Multivariate Statistics and by means of data collection to realize the computer monitoring and optimization control of CCR operation. The content of this article covers mainly the following sections:1) Through analyzing the CCR process and heat transferring principle of CCR furnace based on the Multivariate Statistics Theory, we identified all parameters to be monitored and build up accordingly a data-driven monitoring mathematical model by means of Principle Component Analysis (PCA) for the first time, which can monitor and diagnose the failures in CCR operation with applications of Squared Prediction Error(SPE), Hotelling T2, Principle Component Scores charts, and variates contribution charts. The PCA is improved against the possible non-linearity among variates and the multiscale noise pollutions existing in collected datum. The application result of multiscale linear PCA and multiscale NPCA shows that non-linearity analysis constricts the data effectively, wavelet transform filters the noise pollution in all scales, and avoid the defects of impossible detection of minor deviation and delayed identification of major deviation in datum by PCA.2) Due to the fact that not all observed data obey normal distribution and the shortcoming that PCA can only make use of second order statistic, this article puts forward a kind of Independent Component Analysis (ICA) based monitoring measure for CCR unit. The independent non-gauss value is developed from observed variates by means of non-gauss maximization criteria, which meets not only the non-correlation required by PCA, but also the independence characteristic in the sense of statistics. After we selected independent variates and established relative statistic control limitation, we monitored the CCR operation and the result revealed that there was less faulty alarms and missed alarms than that produced by PCA.3) After the discussion of primary factors causing the dust aggradation and scale formation, and analysis and comparison of domestic and overseas dust amount measurement methods, we present and deduce an on line dust measurement model based on multivariate statistics, which provides a criteria for evaluating the optimized control of CCR furnace.4) An optimization control system based on PCA Controller is built up for the first time, and through effective communications with the original DCS, controlling performance of the original DCS is improved and heat transform of CCR furnace optimized. In the control loop of control oxygen content ,through feedback calculation of heat efficiency to obtain optimal oxygen content in fuel gas and to control oxygen content to reach optimal set point by self-adjust ladderly generalized predictive,control can raise heat efficiency of heating furnace.