Research And Application Of Adaptive Clustering Algorithm Based On Density Peaks

In the background of big data,cluster analysis is one of the important technical means of data mining and has a wide range of applications.Because of the good clustering effect and not constrained by the cluster shape,the clustering analysis method based on the rapid search of density peaks stands out in many clustering algorithms.The key parameters and the center of clustering need to be selected artificial participation,clustering algorithm used to bring a lot of difficulties.In view of the above problems,this paper carried out research on the key parameters adaptive optimization of density peak rapid search algorithm and the method of automatic determination of density peaks,and applied it to cluster analysis based on near infrared spectroscopy.This paper mainly carries out the following three parts of research work:1)Beginning with the basic principle of fast search density peaks clustering algorithm,several common methods of calculating local density are introduced,and the influence of key parameters on local density calculation and clustering performance is analyzed in detail.direction.2)The difficulty in the application of the rapid search algorithm of density peaks lies in how to determine a suitable parameter of the cut-off distance.However,the specific calculation method for this parameter has not yet been clearly defined and usually man-made.This undoubtedly increases the algorithm's Human subjectivity,and with the increase of the number and dimension of data samples,the difficulty also increases.The method of extracting cluster centers from decision-making graph is also vague,which limits the performance of the algorithm.Aiming at the shortcomings of the algorithm,two corresponding improvement measures are put forward:?The method of curve fitting is used to realize the automatic determination of clustering centers;? The self-adaptation of cut-off distance parameter is realized by minimizing the density potential entropy.At the same time,the specific flow chart and steps of the improved algorithm are also given.Finally,several common data sets in the public database are used to verify the improved algorithm,which shows the effectiveness of the improved algorithm.3)Tetrastigma hemsleyanum is a kind of rare Chinese medicinal plants with remarkable curative effect on many kinds of clinical diseases.There are obvious differences in medicinal value of Tetrastigma hemsleyanum green in different areas.The identification of producing areas based on clustering technology is of great significance to supervise the market and protect consumers' interests.In the application of the improved rapid search density peaks clustering algorithm to study the near-infrared spectral data of clover,a total of eight clover green near-infrared spectral data were collected from different regions,and the spectral dimension was up to 1577 dimensional,thus the cutting off distance is difficult to set manually,and the near-infrared spectra in different areas are similar,which brings some difficulties to the accurate extraction of cluster centers.When using the improved algorithm proposed in this paper,the calculation result of this parameter is 0.038,the clustering center is automatically determined as 8,and the final identification accuracy is 100%.It is further demonstrated that the method of adaptively determining the cut-off distance based on the minimization of the density potential entropy is effective.It also shows that the non-linear function can be used to fit the decision-making map accurately to extract the cluster center points.With more and more diversified data,the clustering algorithm based on the fast searching of density peaks has a wide application space because of its fastness and is not constrained by the shape of clusters.