Uncertain Clustering Method And Its Application In Data Streams Processing

In many fields,such as web log analysis,network traffic management,sensor network and network monitoring,data streams have been widely used.While data stream is often uncertain in some degree due to inaccurate measurement,sampling error,outdated data,people's lack of awareness and so on.This uncertainty is challenging the management of data.Traditional clustering algorithms for data streams usually can only deal with accurate data,but they are incompetent to achieve ideal results for the uncertain data streams in some actual applications.This dissertation in the first part mainly is to topic research knowledge background introduction of uncertain clustering analysis and swarm intelligence optimization algorithm.We propose an improved bird colony algorithm.Experiments verify the efficiency of IBSA.We combine the improved bird clustering algorithm into data clustering process,and propose an uncertain data stream clustering algorithm based on bird swarm optimization with feature selection(UDSCBS).UDSCBS algorithm has more features,better clustering effect,higher quality compared with the UMicro algorithm.We also combine clustering algorithm with rough set theory,and propose an uncertain data stream clustering algorithm based on rough sets(CluUR).The experiment verifies the efficiency of CluUR algorithm in dealing with uncertain data streams.The specific contents of this dissertation are as follows:Firstly,the bird swarm algorithm(BSA)has some deficiencies of vigilance behavior and flight behavior in matters of model design.In order to increase targeting ability of bird vigilance behavior,in the selection of individual targets,we choose the best individual to replace random selection methods of the original algorithm in selecting target individual stage,and weighted average of step length is used to overcome the excessive leaping defects of the producer's oversize iteration step in the flight behavior.Experimental results of 12 typical benchmark functions show that the improved algorithm can achieve a better global search ability and an optimized accuracy with better convergence speed.Secondly,this dissertation proposes a new clustering algorithm(UDSCBS)which combines clustering technique with the improved bird swarm algorithm(IBSA).In UDSCBS,the convergence ability of the IBSA algorithm is used to guide the direction of the clustering algorithm and update the cluster center,thus improving the speed and quality of the clustering algorithm.At the same time,the UDSCBS algorithm can feed back the IBSA algorithm with fusion superiority.Experiments with uncertain real data sets verify the efficiency of the selection with redundant features.Finally,this dissertation introduces a new clustering algorithm for uncertain data stream,which defines a new cluster feature of uncertain data streams with rough set theory,and uses the upper and lower approximation of rough set theory to characterize micro clusters.The algorithm judges and selects more suitable micro clusters based on the dissimilarity between the tuple and uses agglomerative hierarchical clustering method to generate new micro clusters in outliers.