Research On Nonlinear Causality Analysis Of Industrial Processes

With ongoing development of industrial production,manufactoiy process becomes more complicated likewise.Nowadays,there are always lots of control loops contained in industrial fields,whose control performance has a lotto do with the quality of industrial production,energy consumption as well as production security.For this purpose,it is of great significance to enforce effective and timely control,which is based on precise acknowledgment of interaction among different loops.This article conducts nonlinear causality analysis for industrial process in the frame of information theory and takes usage of conception of entropy.The main purposes are to clarify causation mechanism of the system and how the oscillation propagates from the source to other loops.1.The kernel concept of the article is Granger Causality.This article brings up a new kind of algorithm for causality analysis which is based on qualitative shape analysis(QSA).QSA accomplishes coarse graining of data through depicting of its variation trend.Besides,QSA is compared with two already existing algorithm KNN and Order,which demonstrates QSA,s effectiveness and veracity.2.Anticipation is a special kind of interaction between variables which states as current value of effect may contain part of future information of cause.As a result,traditional algorithm based on Granger causality can,t detect causal influence accurately.This article put forward a improved algorithm based on KNN with core idea of transforming causality analysis into finding the parent node of the object variable.The improved algorithm is proved to be effective through detection for model with anticipation.3.The combination of QSA with the improved KNN leads to the causality analysis algorithm for complicated industrial process.First,detecting correlation between every two variables,so those inapparent relations can be eliminated.Second,distinguishing between direct and indirect correlations,and the latter can be excluded from the following analysis.Finally,detecting the direction of remaining correlation,i.e.detecting causality,to ascertain the upstreaming of the whole oscillation loop.4.Conducting causality analysis using the combing algorithm on TE process.The results demonstrate that the algorithm brought up by this article can locate oscillation source correctly and find important main causal interaction routes.