Target Pattern Recognition Technology Based On Data Fusion And Its Application

Target pattern recognition is widely used in industrial equipment and system condition monitoring,fault diagnosis,optimal control,cooperative scheduling etc.However,common theory such as principal component analysis,cluster analysis,evidence theory,support vector machine and other theories are essentially in the time domain,which the information within the discrete domain can be processed.Those theories can only achieve target pattern recognition for stationary processes.There are a large number of problems of target pattern recognition of nonstationary processes.When using the traditional method to analyze the signals in time domain,the difference of the signals due to the dynamic time may lead to huge errors,which will lead to misjudgment and miss judgment.Due to frequent changes in the load commands of the power grid and the increase of coal fluctuation,the proportion of time occupied by thermal power units operating in dynamic variable conditions is greatly increased.The idea of this research is to integrate the signal processing method in time domain with the information fusion method in discrete domain to realize the target pattern recognition of dynamic process.The following attempts were made:(1)Information entropy theory,which can reflect the characteristics of signal variation,is applied to AGC(automatic generation control)command pattern recognition.(2)The multiscale filter was used to decompose the signals with inconsistent velocity and selected the signals of the same for D-S fusion.This method was applied to the fault judgment for critical blockage of coal mills.(3)Using the soft-sensing technique to construct the intermediate signals with high static accuracy and high dynamic response speed to make dynamic compensation which is to ensure the original signals are in the same time scale.This method was applied to the safety of heating units' operational pattern recognition aspects during heating season.After mechanism analysis,simulation analysis,operation data analysis and actual debugging,the application results showed that this method can solve some problems on the spot and obtain good actual effect.The subject research has made a meaningful investigation in the target pattern recognition of non-stationary processes.