Evolving Fuzzy Segmentation Algorithm For Multivariate Time Series With Application In Thermal Process

The power industry has always been the pillar industry of our country’s national economy and an important guarantee for promoting the healthy and rapid development of the economy.Therefore,research on the monitoring of thermal process conditions of generator sets has important value and significance.With the universal application of the information system in the generator sets,it has good conditions for long-term acquisition of sets operating data.It also can provide new technical means for data-based operational condition monitoring.However,due to the nonlinearity,strong coupling,large-scale and time-varying of these data,it is difficult to manual analysis.So deep research is needed with the help of data mining theories and methods to achieve the purpose of monitoring the condition of thermal process status of generator sets.The thermal process of typical generator set equipment is taken as the research object in this paper.A Gath-Geva clustering multivariate time series segmentation algorithm based on kernel principal component analysis and evolving theory is proposed from the perspective of time series data segmentation.The main contents can be described as follows:1.In view of the nonlinear,and strong coupling characteristics of thermal process data of generator set equipment,a multivariate time series segmentation algorithm is proposed based on kernel principal component analysis and Gath-Geva fuzzy clustering,which is also called KPCA-GG method.The kernel principal component analysis method is used to extract the nonlinear features between the clustering results obtained by Gath-Geva fuzzy clustering algorithm and construct the nonlinear analysis model space.The distance of the clusters in the model space is used as the criterion for class cluster similarity analysis and merging,thereby improving the accuracy of clustering and achieving the purpose of condition monitoring of such data.2.Aiming at the large scale and time-varying characteristics of the thermal process data of generator set equipment,the evolving clustering theory is integrated into the above KPCA-GG time series segmentation algorithm.And an evolving KPCA-GG segmentation algorithm is proposed to cluster online.The method considers the similarity relationship between the data model at the previous moment and the current time according to the framework theory of evolving clustering.The historical information is used to regularize the modeling of the target cost function.So that the static KPCA-GG clustering algorithm is extended to a dynamic clustering algorithm.The new proposed algorithm can overcome the shortcomings of static algorithms that can only analyze a single time period.It also achieves the purpose of analyzing dynamic segmenting analysis and acquiring results online.And it can process nonlinear data with large data volume and time-varying characteristics better.3.In order to verify the effectiveness of the algorithm,the data collected in the “Yunnan Water and Thermal Power Unit Condition Monitoring Center SCADA” is studied using the method proposed in this paper.The status monitoring experiments are carried out with the data of high-pressure heater equipment as the representative,and the corresponding experimental diagram,experimental data and research results are given.The purpose of these results is to illustrate the advantages of this method for equipment condition monitoring that it is applicable to processing nonlinear data online.4.Using MATLAB’s graphical user interface function to develop a data segment tool based on the algorithm of this paper.The tool can display the processing of system data.It also can visualize the running effect of the algorithm and facilitate parameter adjustment.