Research On Methods Of Multi-Granular Information Fusion Based On Dsmt And Its Applications

Recently,multi-source information fusion has become one of the key technologies of information decision-making system.However,due to the physical limitations of sensor devices,the uncertainty of system operation,and even the unknown dynamic interference of the environment,multi-granularity information fusion has increasingly become a major challenge in the research of information fusion technology.In view of the fact that the theoretical research of information fusion is an effective way to solve the problem of multi-granularity information fusion,this paper carries out in-depth research on the generalized fusion of multi-granularity information based on the framework of DSmT.Besides,Activity Recognition based on WBSNs is also studied.Firstly,from the perspective of uncertainty measurement and fusion rules of multi-granularity information,this paper proposes the concepts of quantitative and qualitative hesitant fuzzy belief mass,associated hesitant fuzzy set theory with belief assignment technology in DSmT.And then we further proposes the related fusion rules under the framework of hesitant fuzzy belief mass.In addition,in view of the premise that the use of fusion rules in DSmT framework is limited to the unique Fo D(Frame of Discriminant),this paper also proposes a novel fusion rule based on rough set theory that can effectively fuse multi-granular evidence sources in different discriminant framework.And the related simulation examples verify the proposed fusion rule;With the increasing number of focal elements in Fo D,the computational complexity of fusion rules for multi-granularity evidence sources greatly limits the practical application of DSmT theory.Here,a novel dimension reduction method based on BF-TOPSIS strategy is proposed.According to the comprehensive evaluation indexes: the mass of focal elements and the dimensions of Martin coding of focal element,the importance of focal element is first judegd based on BF-TOPSIS.And then the operations of preservation and removal of focal elements according to k-l-x strategy are realized.In addition,in order to make the decision results in final fusion result more explanatory,PT(Probabilistic Transformation)needs to be done so as to transform the Non-Bayesian BBA to Bayesian BBA.In this paper,a probabilistic decision-making method based on multi-objective genetic optimization algorithm is proposed.In this method,The assignement of the coarse-grained compound focal element is reasonably allocated to the fine-grained singletons through iterative optimization;To solve the problem of multi-granularity evidence source fusion with unequal reliability,a novel multi-granularity evidence source fusion method based on the self-adaptive BF-TOPSIS is proposed.This method evaluates the reliability of multi-granularity evidence sources from the two aspects: the first one is self-uncertainty within multi-granularity evidence sources and the second one is the conflicts exist between multi-granularity evidence sources.In addition,since the weight of evaluation index itself also affects the evaluations of the reliabilities of evidences,an adaptive evaluation index weight is also given here.Finally,these unequal multi-granularity evidence sources are fused with the discounting fusion strategy and PCR6.The simulation results show that the proposed method has obvious advantages in the fusion of high-conflict or dynamic evidence sources;Aiming at the coping with the uncertainties of information acquisition in WBSNs,this paper uses multi-granularity evidence source fusion strategies in DSmT to recognize human activities in WBSNs.The two novel strategies are introduced: 1)equal reliable multi-granularity evidence source fusion based on KDE and DSmT;2)non-equal reliable multi-granularity evidence source fusion based on K-means clustering strategy,BF-TOPSIS and DSmT.Besides,a novel activity recognition model based on hesitant fuzzy mass and ELM is also proposed to evaluate the efficiency of our proposed hesitant fuzzy mass assignment technique in DSmT.Compared with other state-of-the-art methods,the results show that the proposed activity recognition models have the advantages of higher accuracy in daily activity recognition problems,which provide a new way of thinking for activity recognition in classical WBSNs.