The Research And Application Of Evidence Function Construction And Evidential Reasoning Algorithm

The biggest advantage of the theory of evidence is that it can express the problem of "uncertainty" and "unknown" on the premise of not knowing the prior probability.It provides a reliable method for uncertainty reasoning.Currently,in the field of artificial intelligence has been widely used.However,there is no complete way to obtain the evidence function from the data source,that is,the construction of the basic reliability assignment function and the complex evidence network reasoning method still have many problems to be studied.The main work of this paper is as follows:Firstly,in view of the difficulty of constructing the basic belief assignment function(BBA)in the theory of evidence,a new method for constructing weighted basic confidence assignment function is proposed and applied to multi feature image classification.The method used the recognition accuracy and the posterior probability obtained from multiple logistic regression methods to construct the weight coefficient of the evidence fusion.Afterwards,the weighted BPA is built and classification is done through weighted D-S evidence fusion.Results from the case study demonstrate that the proposed method can overcome the accuracy instability of single-feature based image classification method and improve the classification accuracy.Secondly,aiming at solving the reliability inference problem in the multi-connected knowledge network model,a clique tree propagation algorithm is innovatively applied in this paper.Firstly,the multi-connected network is clustered as a clique tree and the joint belief function is regarded as the main parameter of the cluster nodes;therefore,which facilitates the possibility of reliability inference within the multi-connected knowledge network model.In the process of evident fusion of joint belief function,two new union and intersection methods are introduced to improve the existing DSmT theory,which helps to eliminate the influence of conflicting evident information on other evident variables.Finally,the feasibility of the method is verified by an example.