A Data-Driven Dynamic Data Fusion Method Based on Visibility Graph and Evidence Theory

Dynamic data fusion on time series plays an important role in real applications like target identification. The existing credibility decay models (CDM) may be too subjective for parameters setting and do not make full use of time series information. To address these issues, a new method based on visibility graph and Dempster-Shafer evidence theory are presented in this paper. With the assist of a visibility graph averaging aggregation operator (VGA), a structure revision basic belief assignment (SRBBA) which contains past time information can be obtained. Through this way, the judgment to past data credibility is data-driven without the interference of subjective factors and more reasonable because more time information is considered. Besides, a series of identification applications, including numerical simulation, sensitivity analysis, and practical Iris class identifying are executed to illustrate the efficiency of the proposed method. These applications can show that the proposed method has promising aspects in time series data fusion.