Studies On Methodology Of Multivariate Statistical Process Control With Application To Refinery Process

Speedy development of computer technique has brought ever-growing progresses of the research and application of Multivariate Statistical Process Control (MSPC). In the dissertation, based on the investigation of significant problems of MSPC in the background of refinery process, MSPC techniques are incorporated to the tasks of monitoring of refinery process. Main research work and contributions of this dissertation are as following:Having deeply studied basic theory about Multivariate Statistical Process Control and mastered some basic methods such as Principal Component Analysis and Partial Least Square. Using Exponential Weighted Moving Covariance Plot (EWMA Plot) and MSPC Control Plot (such as SPE-score picture, T2 picture and Principal Component Contribution Plot, etc) to monitoring procedure process, and to examine abnormal conditions in process.There is serious non-linear among process variables in many processing. Though the question can be solved using PLS by eliminating correlation in variables, it can not used to solve process with strong non-linearity effectively because it using linear relation to connect input and output factors. Therefore, studying non-linear PLS methods has excellent application worthiness. Based on the correctitude character of Chebyshev polynomial, this paper brings forward an improved Non-linear PLS and it can be used to represent broader non-linear model.Making use of identities of PCA, an improved Robust PCA method be developed, and it can filter invalidation samples which could not mapped exactly the relation among input and output variables to enhance model building dependability. Using filtered samples to build RBFNN model and compared with model building by principal components.Using RBFNN and Chebyshev PLS methods separately approach non-linear process and comparing results.In the refinery process, it is extremely important to ensure the procedure running safely. And using software for managing can improve the standard and quality of producing managing. In this paper, the author developed software based on Visual Basic, which counts the rate of unqualified producing and writes the result into a report-form. The software can be regarded as the criterion of producing-checking, and it can implement in-line managing successfully.