Research On Clustering Algorithm And Model Of Swarm Intelligence Optimization

Today,with the rapid development of mobile Internet technology,computers have greatly improved their capabilities in high-speed computing,logical judgment,and storage functions.In the face of the huge amount of data generated in fields such as e-commerce and Internet finance,how to mine the useful information contained in the massive data with diverse contents and diverse types has become an urgent problem to be solved under the background of "artificial intelligence".In the classification process,we can find the potential structural relationship of the data between the data without having to classify by the specified classification criteria in advance.Clustering,which can easily obtain the distribution of data,can also view the characteristics of each cluster and further analyze specific cluster sets as a tool.Because of the variety of techniques used in cluster analysis,different conclusions are usually drawn,so cluster analysis techniques with unsupervised learning capabilities have become one of the research hotspots in the field of artificial intelligence.Currently,many clustering algorithms require manual setting of key parameters during algorithm implementation.When we are faced with complex multi-dimensional data,it is difficult for people to find the appropriate global parameters,so how to confirm the appropriate global parameters has become an important research direction to improve the clustering effect.This article takes how to solve such problems as the research direction,uses the advantages of collaboration,distribution,adaptability and robustness of swarm intelligence to improve the firework algorithm,and integrates the clustering method through a new multi-group collaborative intelligence algorithm.Realize the deep mining of data with the same or similar attributes,thus forming a new cluster analysis model.The following 4 parts are the main content and innovation in this article:(1)In view of the fireworks algorithm get into local extreme problem in the searching process easily,this article redefines the original explosion radius formula through dynamic search,and introduces the concept of the minimum explosion radius.In order to make the explosion radius in the algorithm pass the dynamic The calculation of the changed form,and the current number of iterations and the maximum number of iterations are introduced into the formula,while the corresponding physical meaning of the original algorithm is not changed,the fitness value is retained,and the new explosion radius formula is updated in a non-linear decreasing manner.In this way,we can achieve faster global search in the early stage of the algorithm and sufficient local search in the later stage of the algorithm.(2)In order to solve the problem of low precision of fireworks algorithm,this paper adopts the optimal method--tournament selection strategy.The first step is to choose a certain number of individuals,the second step is to choose one of the best in the offspring population,the third step repeated many times the operation,until meet new population size and size to have consistent with the original population size and size,make the results more close to the expected results,achieved the purpose of more accurate precision.(3)The performance of the density peak clustering algorithm is very sensitive to the density estimation,which is the key to selecting a suitable truncation distance(dc).Traditionally,the choice of dc is based on subjective experience.When searching for non-spherical clusters,the algorithm will encounter difficulties in finding suitable dc,especially when a cluster has multiple density peaks,which will cause the clustering effect of the density peak clustering algorithm to be not obvious.In this paper,through the improved firework algorithm,taking advantage of its fast search speed,it is possible to select a suitable truncation distance dc,which enhances the robustness of the algorithm.(4)The density peak clustering is also based on subjective experience when selecting the clustering center.So this paper puts forward a way to automatically determine the clustering center number,which introduces the concept of cluster center weights,and reduces the overall weight of cluster centers Trend,find the highest point of trend change,and determine the number of cluster centers.