Research On Basic Probability Assignment Derivation And Combination In Evidence Theory

Nearly 30 years, with the development of science and technology, the performance of sensor has been improved greatly. In multi-sensor system, the information form is various and the relationship between information is complex. Information fusion is emerged at a historic moment and has got rapid development. Because of the uncertainty of multi-sensor information, reasonable representation and effective combination of uncertain information is a challenging task in information fusion. Evidence theory has its unique advantage in uncertain information processing, but also some problems such as the derivation of basic probability assignment, the applicability of Dempster’s combination rule, and the combination of the conflict evidence.The derivation and combination of basic probability assignment in Evidence Theory are studied, and the contributions of this thesis are listed in three aspects as follows:（1） A method to determine basic probability assignment based on cloud model is proposed, which combines the randomness and fuzziness. Firstly, the normal cloud model of every category under the property is constructed based on the backward cloud generator. Secondly, using the antecedent cloud generator, the certainty expectation of the test samples of every category under this property is generated. Thirdly, a method to measure the similarity of normal cloud models is given, and the maximal similarity of the normal cloud model with the maximal certainty is set as the belief of the universal set. Finally, the certainty is normalized to get the basic probability assignment of every category. The effectiveness of this method is proved by the experiments, and it can determine basic probability assignment in the case of few samples.（2） Changing the way of using the classical conflict coefficient to determine the applicability of Dempster’s combination rule in D-S evidence theory, a new applicable condition of the Dempster’s combination rule is presented. Firstly, the fact that evidence conflict cannot judge the application of Dempster’s combination rule is discoved. Secondly, the reason of the counter-intuitive results of using Dempster’s combination rule is analyzed. Thirdly, a new applicable condition is given, namely, for every element iA of evidence m₁, there is one focal element B_j of the other evidence m₂,A_i∩B_j≠Ф, and m₂（B_j）>ε （ε>0）. The results of experiments demonstrate that the new applicable condition is clear and simple, and provides a good applicable and reasonable indicator.（3） To suppress the counter-intuitive results encountered using the Dempster’s rule of combination, a conflicting evidence combination approach based on the applicable condition of Dempster’s combination rule is putforward. Firstly, the principle how the existing combination method of conflict evidence based on evidential model modification solves the counter-intuitive problem is discussed. Secondly, its disadvantage that the associative property is not satisfied is found. Thirdly, the applicable condition is used for model modification of the original evidences, and Dempster’s combination rule is applied to evidence combination with the associative property satisfied. Some numerical experiments show the efficiency and rationality of the new approach.