Research On Industrial Nonlinear Causality Analysis Based On CCM

Modern industrial process control system generally has hundreds of thousands of control loops.It often has the characteristics of large-scale,high integration,strong coupling,etc.In order to maximize production efficiency and ensure product quality,it is expected that control loops operate at optimal conditions.However,in practice,some loops suffer from different levels of control performance problems,which adversely affect economic efficiency,production environment and safety.Therefore,it is of great importance to conduct accurate causality analysis and identify the root cause of faults as soon as possible.Based on the nonlinear causal analysis method of Convergent Cross Map(CCM),this thesis studies the causal relationships among oscillation loops in industrial control system.The proposed method can accurately capture the oscillation propagation path and identify the root cause of plant-wide oscillations.The main research contents of this study are as follows:1.The existing classical causality analysis methods,such as Granger causality test,traditional CCM,and extended CCM,are compared in depth.The experiment results show that the extended CCM has better performance in accuracy and applicability.Because the traditional CCM method does not consider time-delay characteristics to identify the causality,to tackle this issue,a novel index is proposed,whose effectiveness is verified by experiments.The proposed method can reduce the dependence of CCM on human judgment and improve the automation level of causality analysis.2.A CCM causality analysis method based on denoising and cycle-removed is proposed for the oscillation signals of industrial process control loops.First,based on the three signal decomposition methods of Empirical Mode Decomposition,Variational Mode Decomposition and Singular Spectrum Analysis,a kind of denoising methods based on Detrended Fluctuation Analysis and two kinds of cycle-removed methods are designed.The advantages and disadvantages of the above methods and the characteristics of applicable objects are compared.Then,through simulation comparison experiments,the necessity of denoising and cycle-removed to analyze the causality of the oscillation loops using the CCM is verified.Finally,the three main oscillation loops in the Tennessee Eastman Process are used to diagnose the oscillation propagation paths,verifying the feasibility and accuracy of the method.3.Aiming at the plant-wide oscillations in industrial processes,a CCM causality analysis method based on grouped harmonic detection is proposed.This method uses Multivariate Empirical Mode Decomposition to decompose the oscillation signal into a serial of modes.Then the harmonics can be extracted through the relationships among different group' s frequencies,which indicates loops with nonlinear oscillations.The corresponding causalities can be revealed by CCM causality analysis with the denoising and cycle-removed of these loops.In order to validate the effectiveness,the proposed method is applied to a plant-wide oscillation dataset,which is collected from a hydrogen reformer process of Southeast Asian Refinery.The experiment results show that the propagation paths of nonlinear oscillations are effectively diagnosed.Also,the corresponding nonlinear oscillation source is correctly identified.4.In order to analysis the propagation paths of linear oscillations in industrial cases,a novel CCM causality analysis method based on K-Nearest Neighbor mutual information is proposed.This method selects features by calculating mutual information based on K-Nearest Neighbor between plant-wide oscillation loops.Only the loops with relatively high correlation degree are retained to perform CCM causality analysis with denoising and cycle-removed.The experiments on TE show that,compared with the original CCM,the proposed method outperforms in analyzing propagation paths and locating root cause of linear oscillations.