Fast prediction method in natural gas purification process based on improved KPCA

作者：Shao Yanchao; Xu Liangjun; Hu Yanzhu; Ai Xinbo

来源：International Journal of Earth Sciences and Engineering, 2015, 8(4): 1667-1674.

• Comparative studies
• Kernel principal component analyses (KPCA)
• KKPCA
• KPCA
• Multivariate statistical method
• Nonlinear perturbations
• Prediction accuracy
• Purification process

The research of prediction method in natural gas purification process is necessary in order to improve the product gas quality and ensure production safety. By introducing the kernel theory, kernel principal component analysis (KPCA), the nonlinear multivariate statistical method, is used to evaluate process running status. The comparative study between traditional PCA method and KPCA shows that the KPCA method can detect the influence of small and nonlinear perturbations on the system more sensitively, and shows the better prediction ability than the traditional prediction method. Then method prediction accuracies of different kernel functions, Gaussian kernel, linear kernel and exponential kernel, are discussed in the paper, and the statistics values of T2 and SPE show that the KPCA method based on Gaussian and linear kernel can be predict abnormal condition earlier, and method based on exponential kernel has the better ability of anti-interference performance. In order to train model by large size historical samples, the improved KPCA method-KKPCA is introduced to solve method efficiency problem, which is based on the clustering analysis. The result shows that the new fast prediction method KKPCA can redu

• 出版日期2015
• 单位北京邮电大学