Researches On Fusion Of Incomplete Information Based On Evidence Theory

With the continuous development of technology and economy,human society has made great strides toward informationization.Nowadays it has entered the era of internet and internet of things.It has become the subject of government,business and research institutions that how to effectively fuse the multi-dimensional,multi-level and multi-type information brought by big data to obtain more accurate assessment of objects or targets.In the application of information fusion,information is usually collected by sensors.This is the first and crucial step for effective fusion,as the reliability of collected information determines the accuracy of the final fusion result.However,in practical applications,the observed target is usually unknown and uncertain,and even has the interference information from enemy.So in the complex,uncertain environment,sensors often get incomplete information.In this case,the traditional method of information fusion may lead to inconsistent results,and decisions based on such results may lead to serious consequences.So facing incomplete information,how to carry out effective information fusion will be the main content of this article.Information fusion,as a technique for comprehensively processing data resources,has a variety of methods.Dempster-Shafer evidence theory(D-S theory)has drawn much attention due to its advantages in expressing and processing uncertain information.As a generalization of classical probability theory,D-S theory extends the basic event space in probability theory to its power set,and establishes a basic probability assignment function(BPA)on it.In addition,it provides a combination rule to make it possible to fuse evidence in the presence of information.Therefore,based on D-S theory,this paper explores how to effectively integrate incomplete information.The specific research contents and methods are as follows:(1)To study the problem of how to effectively fuse incomplete information based on KL divergence.We propose a distance-based evidence weighted averaging algorithm for the fusion of incomplete information.We firstly define the method to measure the difference between BPAs based on KL divergence,which also can be regarded as a measure of the distance between evidences.Then,a weight generation method of different evidences is proposed based on the above definition,which can improve the classical evidence fusion rule,so as to achieve the purpose of effective fusion of incomplete information.(2)To study how to effectively deal with incomplete information fusion problem by improving the quality of information sources from the perspective of entropy.In this section,the concept of positive ideal basic probability assignment and the negative ideal basic probability assignment are firstly defined based on the belief entropy proposed by Jirou(?)ek,then the concept of evidence reliability closeness index is proposed to measure the reliability of different evidence bodies,and the algorithm to fuse incomplete information effectively is presented finally.By constructing numerical examples,the proposed algorithms are compared with other classical incomplete information fusion methods.The comparison results show that the proposed algorithm has higher accuracy and better convergence.In order to further highlight the superiority of the defined algorithms,we conducted a set of comparative experiments based on open dataset,and achieved the contrast effect by constructing different types of incomplete information.The experimental results show that our proposed algorithms have a wider scope of applications,and the effective fusion results can still be achieved under complex and uncertain environment.And the proposed methods can converge faster.