Research On Data-driven BPA Generation And Conflict Measurement Methods

With the advent of the information age,multi-sensor information fusion technology has attracted more and more attention.How to obtain effective information in the complex uncertain information becomes very important.As a method for dealing uncertain information,D-S evidence theory can make decisions by fusing multi-source information,which is has been widely applied in the field of information fusion.However,it also has some problems.For example,in D-S evidence theory,the conflict factor k can not indicates the degree of conflict accurately.In addition,how to determine the basic probability assignment(BPA)based on the obtained uncertainty information is also an important problem to be solved when D-S evidence theory is applied in practice.To address the above issues,this paper proposes two data-driven methods for the determination of BPA and a new method for the measurement of evidence conflict.The main research contents are as follows:(1)Based on the probability distribution of the samples,a new BPA determination method is proposed.Firstly,the probability distribution of the samples can be obtained by fitting the known samples.Then the distance between the test sample and the interval is calculated based on the proposed distance formula under different probability distributions.Next,the similarity is calculated by the obtained distance.Finally,the similarity is normalized to determine BPA.(2)Based on the maximum entropy model,a new BPA determination method is proposed,which constructs the model only based on the known information without making assumptions about the unknown information.Moreover,based on the maximum entropy model,Laplacian smoothing and interval number model are added to optimize the maximum entropy model to solve the problem of sparse samples.BPA can be determined quickly and efficiently by constructing the model.Several simulation experiments are used to illustrate the method,which is effective than others.(3)For the problem that D-S evidence theory cannot measure the degree of conflict,this paper proposes a method to measure evidence conflict based on Bray-Curtis dissimilarity.Firstly,the belief dissimilarity formula is proposed based on Bray-Curtis dissimilarity.Then,the composite proposition in the original evidence is transformed into a single subset proposition by introducing the Pignistic Probability Transformation.Finally,we can calculate the degree of the conflict between two pieces of evidence based on the belief dissimilarity formula to get the final result.Several classical example simulations are used to illustrate the effectiveness and superiority of the proposed method from two aspects of theoretical derivation.Moreover,the effectiveness of the proposed method in evidence combination is simply verified.