With the continuous development of science and technology,significant changes have taken place in the industrial production process.The development level of the modern process industry as a national pillar industry not only represents the country's competitive strength and social and economic development level,but also is closely related to the people's living standards.At the same time,the various links of the process industry are usually interrelated,and any deviation or abnormality of any link will affect other related parts and even cause a series of serious production accidents.Therefore,how to ensure the safe and stable operation of the process industry has become an urgent engineering and technical problem to be solved.As an effective solution,process monitoring has received more and more attention from scientists and technicians,and has been further developed.The independent component analysis multivariate statistical method is adopted here,and the related theoretical and applied research is carried out according to the actual needs of process industry process monitoring.The main work is as follows:First of all,according to the characteristics of close correlation between process industry process variables,independent component analysis(ICA)is used to analyze high-order statistical information of simulated variables to achieve accurate and effective feature extraction and processing.Among them,the FastICA algorithm based on negative entropy is systematically discussed,and the independent component decomposition simulation experiment of the mixed signal under no noise and noisy conditions proves the good signal separation ability of the method.Secondly,in view of the difficulty that traditional methods cannot select critical and reliable independent components,an evaluation method based on cluster analysis is used to effectively select the separated independent components,and accordingly two types of statistical monitoring indicators are established for different selection results.Realizing the accurate description of the system status provides a more accurate objective basis for process monitoring.Thirdly,for the problem of setting control limits of statistics,a non-parametric kernel density estimation method is used to estimate the probability density function of each statistic,and then calculate their respective reasonable control limits.Among them,for the two main influencing factors of kernel function and window width parameter in kernel density estimation,the kernel function uses Gaussian kernel function;window width parameter uses genetic algorithm for global optimization to obtain the optimal window width that meets the actual problem Into the kernel density estimation formula,the probability density function of the monitoring statistical index is obtained.Finally,a complete process monitoring process based on ICA is given.Through the case study of multiple faults in the TE process,the feasibility and effectiveness of the process monitoring method based on ICA are verified;and based on the above research results,the corresponding design and development The ICA process monitoring software can provide basic software tools for the application of this method.Through the above theory and application research,the relevant technical theory based on ICA-based multivariate statistical process monitoring is further improved and enriched,which provides new options for ensuring the safe and stable operation of process industry processes.At the same time,it also prospects the next step of research work. |