Research And Implementation Of Fast Classification Algorithm For Partial Discharge Signals

Transformers are an essential means of power transmission,and transformer outages can lead to widespread power system outages.Condition monitoring of power equipment provides the opportunity for continuous operational service,early detection of problems and possible remedial measures,thus increasing the life expectancy of power transformers.Different partial discharges exist,and other types of eruptions correspond to various equipment maintenance measures.Phase Resolved Partial Discharge(PRPD)mapping is used to determine different types of discharges.Still,when there are multiple mixed partial discharge signals,the PRPD mapping overlaps,making it impossible to identify the partial discharge type accurately.How to separate various mixed partial discharge signals is of great significance for subsequent data processing.This thesis studies the online separation technology of multiple hybrid partial discharges in power equipment and uses the clustering algorithm to separate multiple partial discharge signals.According to the requirements of practical applications,to separate multiple mixed partial discharge signals,it is necessary to meet the criteria of minimizing human intervention in the clustering process.The number of microclusters cannot be determined in advance,the efficiency of clustering,and the algorithm’s portability.This thesis proposes an adaptive parametric fuzzy C-means clustering analysis method for multiple hybrid partial discharges.The method takes the method of local density estimation to initialize the clustering centres,evaluates the clustering validity index to select the optimal number of clustering categories,and then separates the multiple mixed partial discharge signals by using the fuzzy C-means.This thesis finds that the multiple mixed local discharge signals change continuously when using UHF sensors for local discharge pulse acquisition.It is impossible to adjust the clustering parameters adaptively according to the real-time data characteristics.In order to maintain the entire IFT period of the local discharge pulse signal,a data stream clustering algorithm is adopted for separation.In this thesis,we propose an online data stream clustering method with adaptive parameters,which adopts the concept of the natural neighbourhood to create K-dimension(KD)trees to improve the efficiency of querying the nearest neighbours and adapt the features of stream data to get the neighbourhood radius and area density,so as to perform local search and form clusters to realize the online separation of multiple mixed local discharge signals.Finally,the accuracy and stability of the clustering algorithm proposed in this thesis were tested by actual data sets.The experiments show that the clustering algorithm proposed in this thesis has an average purity of 98.75%,an accuracy of 97.23% and an F-Measure value of 98.14% in the multiple mixed partial discharge signals data set.The clustering algorithm proposed in this thesis is enough to separate multiple mixed partial discharge signals accurately and has a significant reference value for multiple mixed partial discharge separations.