Research On Clustering Algorithm Based On Manifold Distance And Its Application In Aurora Classification

The cluster analysis is one of the effective means and main methods for human exploring things that has not obvious relationship. Unlike classification method based on supervised learning, clustering algorithm does not have the training set and any prior knowledge of the data set, just according to the similarity between feature vectors. In recent years, the spectral clustering algorithm used in this paper get more attention by many researchers in all kinds of the field. It's a new clustering analysis algorithm. Spectral clustering algorithm has its own unique advantage. For example, it can be clustered in any irregular shape of the sample space, but also be obtained the optimal solution in the global.so, it has more advantageous to solve the realistic problem.The spectral clustering algorithm process is as follows:Firstly, I established a samples of similar connection diagram, and the similarity measure to construct a sample data set the similarity matrix of W. secondly, I constructed and calculated the normalized Laplacian L and the largest eigenvalue and eigenvector. And then, to choose one or more of the eigenvector matrix L using the k-means or other data points of different clustering algorithm. In this paper, I prefer to use the clustering algorithm of the similarity measure as the breakthrough point to improve the traditional similarity measure. The traditional spectral clustering algorithm (NJW-SC) of the similarity measure based on Euclidean distance replace based on the similarity measure of popular distance. On the basis the object set and the sample clustering can be clustered. After I set the experimental comparison with the new algorithm and K-means algorithm, traditional spectral clustering algorithm (NJW-SC), the fuzzy clustering algorithm (FCM) on artificial data set, it can be concluded that the new algorithm has been achieved good results in the convex shape of the data sets and on the global consistency. On UCI data sets, I tried to use the artificial labeling evaluation index F-measure numerical calculation to carry out on the clustering quality. After that, spectral clustering algorithm is applied in the classification of the aurora, validation of spectral clustering algorithm also can get very good application in the aurora classification.The main content of this paper includes the following several aspects:1?The classification of the existing clustering algorithms, as well as each kind of classification of the main ideas and the commonly used algorithm.2?Introduce the definition of clustering, clustering criterion, and the clustering process of basic concept clustering algorithm, also expounds the common clustering algorithm, such as K-means?FCM?AP etc.3?Spectral clustering algorithm are introduced in the rule of graph theory basis, the division of spectral and the spectral clustering algorithm classes are divided into according to the spectral classification criterion.4?On the basis of the traditional spectral clustering algorithm (NJW-SC), the similarity measure based on Euclidean distance replaced the similarity measurement based on popular distance.I proposed new clustering algorithm, and experimented on the different data sets.5?The popular distance spectral clustering algorithm is used in the practical problems of the classification of the aurora.1 found in the practical problems in the clustering result is superior to other algorithms.6?summarized the whole thesis and put forward the problems in the process of experiment, and it can be improved in the future.