| Safety has become a considerably major issue that cannot be ignored in modern industrial production,it is important to improve production safety that timely detection of the root variable when the pivot process variable changes abnormally and even alarms.As an effective alarm analysis technology,alarm root-cause analysis is attracting more and more attention in modern industrial alarm systems.In order to provide reliable technical support for modern alarm root-cause analysis strategies,this thesis associates the alarm root-cause with the difference of the multivariate trend characteristics and proposes a method for the alarm root-cause analyzing for thermal power generation units based on trend feature clustering,the paper adopts a brand new idea to analyze the alarm root-cause for thermal power generation units,and the key technologies used in the alarm root-cause analysis are optimized from the following three aspects.1.An effective trend extraction method is used to perform piecewise linear representation of the operating data,and this thesis applies a newest method about determining the optimal number of segments to segment the data reasonably in the process of piecewise linear representation of the operating data,thereby the main change characteristics of abnormal operating data is extracted accurately to provide more valuable data information for subsequent alarm root-cause analysis.2.A method for determining the correlation of variables based on goodness of fit is proposed,which can accurately select the associated variables with strong correlation with the pivot process variables,simplify the work intensity of screening the associated variables,and efficiently provide reliable associated variables for the subsequent alarm root-cause analysis.3.A new density-based clustering algorithm is applied to the feature clustering of changing alarm data abnormally.In terms of the selection of clustering centers,manual selection is changed to automatic selection.In terms of determining the optimal number of clusters,a method determining the "knee" based on the L-shaped curve is introduced to determine the optimal number of clusters,which reduces the amount of calculation and achieves a very applicable clustering effect,thus providing reliable technical support for alarm root-cause analysis.Numerical simulations and industrial cases are given in order to verify the research results.The trend feature clustering method applied in the alarm root-cause analysis for thermal power generation units has achieved satisfactory result,the effectiveness of the industrial cases provides the further guarantee for the future application of the proposed clustering method based on trend feature. |
