Intelligent Information Fusion for Conflicting Evidence Using Reinforcement Learning and Dempster-Shafer Theory

Abstract

Multi-sensor information fusion is a information fusion technology to improve system performance, which plays a key role in actual production and application. Dempster-Shafer theory (DST) can achieve information fusion due to the effectiveness of processing uncertain information without prior probabilities. However, when the evidence is conflicting, it may produce counter-intuitive judgments. In addition, the existing methods need to obtain all sensors information to deal with the conflict, which make them impossible to realize real-time fusion of online information in practice. In order to solve the above problems, we propose an intelligent information fusion method based on the reinforcement learning (RL) and DST, named the DST-RL method. Specifically, the introduction of artificial intelligence technology to realize adaptive conflict processing, which can achieve effective removal of inaccuracy information and avoid the inaccuracy caused by human intervention. Then the Dempster's combination rule (DCR) is adopted to achieve effective fusion of multi-sensor information. On the one hand, the DST-RL method can realize efficient multi-sensor information fusion. On the other hand, it can reduce the complexity of the system when the amount of information is large. Numerical example and application simulation show that our proposed intelligent information fusion method can achieve significant performance superiority in processing online conflicting information.