Research On Density Peak Clustering Algorith

Clustering is an unsupervised method in the field of data mining that solves various problems.It classifies messy objects into different clusters based on certain features,with the aim of making the research objects within each cluster as similar as possible,and as dissimilar as possible between different clusters.The density peak clustering algorithm can identify data of arbitrary shapes,intuitively determine the number of clusters,and easily detect noise points,demonstrating excellent robustness.Function optimization is a classical application area of optimization algorithms and a common optimization problem for performance evaluation of optimization algorithms.Researchers have constructed different types and complex forms of functions: continuous and discrete functions,single-peaked and multi-peaked functions,etc.For some multi-peaked,multiobjective and non-linear function optimization problems,many optimization algorithms are difficult to achieve good results.Based on this problem,the optimization algorithm is improved using the density peak clustering algorithm,and the improved optimization algorithm is applied to function optimization as a way to improve the optimization effect in complex and variable optimization problems.The purpose of this thesis is to improve the density peak clustering algorithm using the siphon equilibrium method,and to combine the improved algorithm with the particle swarm algorithm and the multi-objective particle swarm algorithm and apply them to the UCI data set,the single objective function optimization problem and the multi-objective function optimization problem,respectively.The specific research content is as follows:(1)Aiming at the problem that the density peak clustering algorithm needs to manually select the cluster center,a density peak clustering algorithm based on the improved siphon effect is proposed(IDPC).This method uses the improved siphon effect,removes the weight of the largest cluster center after sorting,selects a relative balance point,and completes the clustering of various shapes of research objects.Finally,the method is applied to the UCI data set to realize the automatic selection of cluster centers and adaptively process data sets with low dimensions and various distribution shapes.(2)A particle swarm algorithm based on a new learning strategy of IDPC(IDPC-PSO)is proposed to address the problems of low population diversity and low global search capability of the particle swarm algorithm.The method uses IDPC to make the particle swarm adaptively separate into several subswarms,tests the inertia coefficients suitable for the algorithm,and designs a new learning strategy for the ordinary particles of each subpopulation,decomposing the whole complex search process into several simple ones,simplifying the problem while increasing the diversity of particles.The global optimum is selected by comparing the fitness values of each locally optimal particle.The method is finally applied to single-objective function optimization problems such as Sphere,which enhances the population diversity and global search capability of the particle swarm algorithm and effectively prevents the algorithm from falling into a local optimum.(3)A multi-objective particle swarm algorithm based on a new learning strategy of IDPC(IMOPSO)is proposed to address the problem of selecting leader particles for multi-objective optimization problems.First,the method utilizes each dimension of the IDPC partition function to make the particles obtain diverse search information.Second,the particles in the current solution set are sorted in descending order according to their congestion distances,and a high congestion distance particle is designated as the leader particle for the ordinary particles in several subswarms respectively,and the average information of the leader particles of other subpopulations is used to design an update method for the leader particles of the current subpopulation to enhance the exploration of less exploration of less crowded regions.Finally,a simple local search scheme is invoked to make the new non-dominated solution replace the old non-dominated solution.Finally,the method is applied to multi-objective optimization functions such as Binh and Korn to obtain a large number of new nondominated solutions.In summary,the IDPC algorithm proposed in this paper has been applied in the UCI dataset.The results show that the method can effectively use the density information and center offset distance of the data,and has good cluster number selection ability and clustering performance.outperforms similarity clustering methods.The combination of IDPC algorithm and particle swarm optimization algorithm has been applied to the single-objective optimization function.The results show that this method can effectively improve the space search ability of particles,and has fast and efficient optimization performance,which is better than similar optimization algorithms.The combination of IDPC algorithm and multi-objective particle swarm optimization algorithm has been applied to the multi-objective optimization function.The results show that this method can effectively guide the learning of leader particles,and has good performance in exploring new non-dominated solutions in unknown areas,which is better than similar to optimization algorithms.