A novel clustering algorithm for typical background harmonic mode identification

With the integration of special harmonic sources such as wind power and photovoltaic power into the grid, new power electronic systems can cause problems with imperfect harmonic analysis and evaluation systems. In response to the above, a clustering algorithm-based background harmonic voltage division method is proposed to provide a basis for the comprehensive background harmonic measurement and evaluation scheme for different operation modes. First, the background harmonic voltage data are preprocessed and features are extracted for dimensionality reduction, then clustered using the Calinski Harabasz-DensityMax (CH-DensityMax) clustering algorithm that determines the number of clusters and optimizes the initial clustering centers to identify the typical patterns of background harmonics. In tests of real arithmetic cases, the average CH value, average SC value and clustering time of K-means clustering algorithm with improved clustering number and initial clustering center are better than those of K-means + + clustering and fuzzy C-means clustering DBSCAN clustering algorithm.