Research On Data Fusion Technology Based On DS Evidence Theory

With the arrival of the big data era,data fusion technology has attracted much attention.In multi-sensor data fusion systems,the data provided by each sensor often has issues such as incompleteness and imprecision,resulting in a large amount of uncertainty in the data processing process.DS(Dempster Shafer)evidence theory is an important decision level data fusion method that has the advantage of handling uncertain information.However,in practical applications,there are still some problems with DS evidence theory,including the construction of Basic Probability Assignment(BPA)functions,conflict evidence fusion,and high computational complexity.This article proposes corresponding solutions to the main problems in DS evidence theory,and the main research content is as follows:(1)Aiming at the problem that the DS evidence theory cannot convert objective data into BPA in recognition frameworks,a BPA generation method based on fuzzy sets and Gaussian models is proposed in this paper.This method first uses a Gaussian membership function to establish a Gaussian model of the original data,and then uses fuzzy set theory to match the tested sample with the established Gaussian model,and obtain the corresponding matching values.Next,normalize these matching values to obtain the required BPA.All BPAs are fused using the combination rules of DS evidence theory to obtain the final fusion result.Then use the Pignistic probability transformation model to convert this result into probability and make the correct decision based on it.Through experimental analysis,it can be seen that our method has achieved good classification results in three commonly used UCI datasets,and the classification accuracy is better than other comparative methods,indicating the effectiveness and practicality of our method.(2)Aiming at the problem that the combination rule of DS evidence theory may cause the combination result to violate the common sense when dealing with conflicting evidence,an improved DS evidence theory fusion method based on cosine similarity and zero element correction is proposed in this paper.This method uses cosine similarity to calculate the similarity between different evidence,assigning higher weights to more reliable evidence,thereby improving the accuracy and reliability of the fusion results.In order to avoid the problem of one vote veto in evidence synthesis,the method adopts the strategy of zero element correction to the weighted evidence body.This processing method ensures that even after excluding evidence bodies with significant errors,relatively accurate fusion results can still be obtained.Finally,the effectiveness and rationality of the method proposed in this paper were demonstrated through the analysis of four conflicting paradoxes and multiple sets of evidence body examples.In addition,this method has also achieved good performance in fault diagnosis experiments on the bearing dataset of Case Western Reserve University.(3)Aiming at the problem that the computational complexity of DS evidence theory increases exponentially as the number of focal elements in the recognition framework increases,an approximate calculation method based on energy function for DS evidence theory of discarding focal element redistribution strategy is proposed in this paper.The traditional approximation methods usually only consider the focal element reliability level,while this method selects the energy function as the standard for measuring focal elements,while considering the influence of focal element reliability and focal element modulus on computational complexity.This method adopts a strategy of discarding focal elements and redistributing them,dividing the original focal elements into discarding focal elements and retaining focal elements.Based on the complete set relationship between focal elements,the reliability of discarding focal elements is redistributed to ensure that useful information in discarding focal elements can be retained.This method successfully reduces the computational complexity during the fusion process and ensures the accuracy of the fusion results.Finally,the effectiveness and rationality of the proposed method were verified through detailed numerical analysis.