Research On Evidence Distance Algorithm And Application For Multi-source Information Fusion

In recent years,the process of social informatization has accelerated,various kinds of information have continuously emerged,and multi-source information fusion technology has also been rapidly developed.How to effectively integrate different levels,different types,and different dimensions of data from multiple information sources has attracted the attention of various industries and different fields.In practical applications,a system composed of multiple sensors may have noise due to its own reasons and the influence of external environmental factors in the data captured by scattered sensors,which makes the acquired information uncertain and conflicting.How to effectively measure the conflicts between multiple sources of information so as to achieve reasonable and credible data fusion is one of the hotspots in domestic and foreign research.Because the signals conveyed by different information sources are different,it is of great significance to construct an appropriate conflict measurement based on actual problems,and this will also be the main research content of this article.Dempster-Shafer evidence theory has obvious advantages in representing uncertain information,because it can effectively model incomplete information and has attracted the attention of researchers.Although the Dempster-Shafer evidence theory has many excellent qualities,it will get counter-intuitive results when it integrates highly conflicting evidence.In order to solve this problem,there are two mainstream solutions: the first is to modify the evidence fusion rules,and the second is to preprocess the evidence,mainly to assign weights according to the credibility of the evidence body before merging the evidence,thereby reducing the impact of controversial evidence on the final fusion result.In order to measure the degree of conflict between evidence bodies,this article will propose two evidence distance measures for different data types.We prove that the proposed method satisfies the properties of distance and discuss the properties of the proposed method.Numerical examples of multi-source information fusion prove that the proposed distance has strong sensitivity and can effectively measure the similarity between evidence bodies.The comparison with the existing distance function shows that the method proposed in this paper can effectively overcome the shortcomings of the existing methods,and it is more robust and accurate in characterizing similarity.The main contents are as follows:(1)Proposed the distance of evidence used to measure the classical belief function.By analyzing the causes of similar collision problems caused by the Jousselme distance function in measuring the difference between focal elements,we propose a new distance measure,which is modeled on the basis of the Jousselme distance function framework.This measure aims at the change of the number of elements in a single set,and describes the similarity between different focal elements through the numerical ratio of intersection and union.We proved that the distance satisfies the distance theorem and can effectively represent the conflict between the evidence bodies.In the case of multi-source information fusion,our proposed method shows obvious advantages.(2)Proposed the distance of evidence used to measure the ordered belief function.Aiming at the situation when the belief function is discrete values in the metric space,we propose a measurement that can reflect the physical distance relationship between elements.Hausdorff distance is often used to measure the difference between point sets.According to its basic idea,we define a similarity matrix to quantify the distance between focal elements.This similarity matrix can characterize the difference in the distribution of elements in a continuous space.Even if the focal elements do not overlap,the distance we propose will still vary with the physical distance.We use normalization to eliminate the influence of data magnitude,and at the same time we prove that the measure is equally effective for interval sets,so the new distance can be used in a wider range.