Information Fusion Method Based On Interval Evidence Theory And Its Applications

As an information fusion method, classic Dempster-Shafer evidence theory cancombine multisource information to reduce the uncertainty and yield more accurateresults. Applying classic D-S theory in information fusion should construct firstly theframe of discernment of corresponding system, and obtain corresponding evidence(Basic Probability Assignment, BPAs) from several information resources of system,combining the evidence and making decision according to the combined results. Theclassic D-S theory requires precise (crisp) BPAs, However, the precise BPA is toocoarse and incomplete to measure uncertain information, and it will miss usefulinformation. In recent years, as extension of classic D-S theory, interval evidencerepresents uncertain information as interval BPA (IBPA), i.e, interval evidence. IBPAwill contain more information than precise BPA and meet human’s generalunderstandings and conceptions. Moreover, more accurate decision can be made byusing the fused IBPA.This paper introduces basic conceps of classic D-S evidence theory and intervalevidence, and investigates the issues of construction of interval evidence andcombination of conflicting interval evidence. The main contributions are as follow:1. Introduce basic concepts of classic D-S evidence theory and interval evidence,analyze and summarize some drawbacks of the existing approaches of intervalevidence in detail.2. Based on modified Latin Hypercube Sampling (MLHS) technique, a newmethod of constructing interval evidence is proposed. This method applies the thoughtof uniform sampling of MLHS to handle interval information in order to constructinterval evidence. Finally, the diagnosis examples of machine rotor show theeffectiveness of this proposed method.3. Because the existing normalization approaches often leads to the loss ofavailable information contained in original interval evidence, a new approach ispresented to realize global normalization of interval evidence. This approach globallynormalizes the valid interval evidence, and makes the best use of information oforiginal interval evidence. Compared with the existing approach, this approach ismore similar to the corresponding original IBPA, so has more information for combination. Numerical examples are provided to prove the effectiveness of thismethod.4. Based on the similarity measure of evidence, a new method for combiningconflicting interval evidence is provided. The credibility degrees of interval evidencecan be got by using the defined extended Pignistic probability function and thenormalized Euclidean distance of interval-valued fuzzy sets. Based on the credibilitydegrees, new interval evidence can be obtained by modified and weightedly averagingthe original interval evidence. Based on Demspter interval evidence combination rule,the fusion result can be obtained by combining the new interval evidence. Theproposed method can effectively eliminate the effect of highly conflicting intervalevidence in combination so as to reduce the width of combined interval evidence.Therefore the uncertainty of decision-making can be decreased. Numerical examplesare provided to prove the effectiveness of this method.