Improved Method Of Conflictive Evidence Combination Based On The Markov Chain

In recent years,Markov chain theory has been widely used in data fusion and information prediction.By analyzing the existing method of conflict evidence combination,we propose an improved method of conflict evidence combination method based on the Markov chain.In general,the methods of conflict evidence combination only consider the influence of conflict coefficient or evidential distance on evidence,and only use conflict coefficient or evidential distance to calculate the weight of evidence,making the calculation of weight is not accurate enough.Based on the combination of conflict coefficient and evidential distance,we use the conflict coefficient matrix and evidence distance matrix to calculate the credibility of the evidence,then,propose a new weighting method.This method takes into full consideration the relativity between evidence and evidence,evidence and coke,which makes the weight of the evidence with small conflict and small conflict to the majority,and improves the weight of evidence.The new evidence weight is used as the discount factor to revise the evidence structure,then,using the Murphy combination method to combine the evidence representation,which are produced by Markov chain transition probability matrix.Through the analysis of the experimental results,the evidence input order of the best synthesis result of the evidence synchronization is obtained.We proposed a conflict evidence optimization method by considering the interference of the high conflict evidence to the accuracy of the synthetic result when the evidences are sequential and asynchronous,which use the average evidence to replace the conflict evidence.This method combines the advantages of model modification and rule modification,which effectively solves the problem of one-vote veto and robustness in conflict evidence combination.Moreover,it can reduce the decision-making risk.Finally,the contrastive method is compared with the proposed method.Experiments show that the proposed method has high recognition precision and fast convergence speed.