An improved approach to generate generalized basic probability assignment based on fuzzy sets in the open world and its application in multi-source information fusion

Abstract

The generalized evidence theory (GET) is an efficient mathematical methodology to deal with multi-source information fusion problems. The GET has the capability of handling uncertain problems even in the open world. In real world applications, some noise or other disturbance often makes the multi-source information have uncertainty. Thus, how to reliably generate the generalized basic probability assignment (GBPA) is a key problem of GET, especially under the noisy environment. Therefore, in this paper, we propose a novel approach to generate GBPA with high robustness by using a cluster method. In this way, the proposed model has the ability to correctly identify the target even under a noisy environment. In particular, the k-means++ algorithm based on triangular fuzzy number is applied to build the GBPA generation model. According to the proposed GBPA generation model, the related similarity degree is calculated for each test instance. After resolving the existing conflicts, the final GBPAs are obtained by using the generalized combination rule. To demonstrate the effectiveness of the proposed method, we compare the proposed approach with related work in the applications of classification and fault diagnosis problems, respectively. Through experimental analysis, it is verified that the proposed approach has the best robustness to generate the GBPAs and maintain a high recognition rate under both noisy and noiseless environments.