Research Of Semantic Integration In Heterogeneous Database Based On Nerual Network

With the rapid development of IT industry, a lot of information management system has built up in the transport system,and large amounts of historical data are accumulated. But with the increasing requirements of comprehensive information usage,those simple information management systems become information islands,on which data sharing become more and more difficult.The subject backgroud of this paper comes from Dalian Municipal CommunicationsBureau data center building. Database heterogeneousness of mainly reflected in twoaspects:one is different database management systems,the other is different semantics.Finding the corresponding semantic objects is the most important issue in heterogeneousdatabase integration domain. Firstly,the present main techniques for heterogeneous databaseintegration are surveyed comprehensively,and the existing problems of present semanticintegration techniques are also studied,and the feasibility to resolve attribute matchingproblems are analyzed using neural network;secondly,the process of semantic integration inheterogeneous database based on nerual network is described: we extract attributes characterswhich includes data schema and data content statistic from database, and then normalize theattributes characters;and then we use the SOM model to classify the attributes characters; andlast we establish BP model by training the sample database to form the matching rules for thesemantic matching between heterogeneous databases.And because BP network with gradientdescent has a low convergence speed and immersing local minimization and PSO also has thedefect of premature phenomena,in the paper gives a method to improve PSO whichconsidered the idea that the organisms have the phenomena of escaping from the originalpopulation when they find the survival density is too high to live.The new algorithms forattribute matching are proposed,and the experimental results show our proposed approach canimprove the attribute matching accuracy and decrease the training time obviously.A system of semantic integration in heterogeneous database based on nerual network is established. And a new algorithm improved PSO training BP for attribute matching is proposed,and the experimental results show the algorithms can improve the precision and recall obviously.