Clustering Method And Application Of Stellar Spectrum Based On Attribute Weighting

At present,clustering technology is more and more widely used in big data processing and analysis.In this paper,the LAMOST stellar spectral data analysis is used as the research background.The two classical partition-based clustering methods(K-means and K-medoids)are optimized to address the problems that the importance of attributes in distance measurement has not been fully considered,the initial center is randomly selected and processing massive highdimensional stellar spectral data are expensive in time and space.The main research contents are as follows:(1)A new clustering algorithm based on attribute weighting is proposed.The traditional partition-based clustering algorithm fails to distinguish the importance degree of attribute features effectively.On the basis of the idea that the contribution rate of each attribute feature to clustering is not equal,this paper proposes an improved clustering that can quantify the importance of attributes.Firstly,a discrete function is defined to distinguish the importance of attributes by their dispersion degree.Then,the similarity distance of each attribute is calculated according to the dispersion value of each attribute,and the sum of the distances of all attributes is used as the basis for judging the similarity between data objects.Finally,clustering is performed based on the idea of K-means algorithm.Theoretical analysis and related experiments on UCI and LAMOST stellar spectral datasets show that the proposed algorithm can reduce the number of iterations in the clustering process and improve the accuracy of clustering.(2)An adaptive attribute weighting clustering algorithm is proposed.Aiming at the problems of random selection of initial centers and insufficient flexibility of fixed attribute weighting in two typical partition-based clustering algorithms,an adaptive attribute weighting clustering algorithm is proposed.Firstly,the ratio of global and local attribute dispersion is defined as the measurement basis of attribute importance.Then,a method of selecting the initial clustering centers by combining density and distance is proposed,which is based on the idea that the initial clustering centers should satisfy the requirements of greater density within the same radius and relatively far distance between the centers.Finally,the "elbow method" is optimized by introducing the adjustment parameters and the optimal cluster number K is determined.The results of UCI and LAMOST stellar spectral data test show that the clustering results of the proposed algorithm are closer to the real partition situation.(3)A prototype system of stellar spectral data cluster analysis is designed.First of all,since the idea of data reduction,this paper constructs a stellar spectral data feature extraction model based on Haar wavelet transform.The model can extract the main features of the spectral data and be used to reconstruct the spectral line feature datasets.Secondly,on the basis of the above research,a prototype system of stellar spectral clustering analysis is designed and implemented based on algorithm(2)to improve the efficiency of human-computer interaction.Finally,stellar spectral datasets test verifies that the clustering results of the system conform to the MK classification law.The development of the system is helpful to the cluster analysis of stellar spectral data and astronomical spectral data mining,which provides a strong support for further analysis of astronomical big data.