Research On Representation And Modulation Of Uncertainty In Multi-source Data Fusion

The demand of perception has been rising nowadays due to the development of information and internet technology. Perception is the foundation of getting information and knowledge of the world, and it has been a hot topic how to integrate and process the perception data, to acquire pure information and make efficient decision in many areas. Multi-source and multi-dimensional are important features in big data environment, thus Multi-source Data Fusion(MDF) has been a popular topic in Big Data and Internet context, which is defined as synthesizing multiple source data into a consistent and advanced result. The essence of information processing is to model and eliminate uncertainty, however, it has been a big problem that there are so much uncertainty and unreliability due to the complex perception scenarios and unreliable source, as a result, the qualify of application and serves is far away from requirement. This research focused on analyzing the relationship of MDF and uncertainty and the framework of uncertainty modeling. Follows are the main works:Firstly, this paper collected uncertainty theory both from classical and latest research and analyzed the quantization of total uncertainty that considers different types of uncertainty and different aspects.Secondly, the relationship of MDF and uncertainty was studied. Its availability can be proved by entropy theory that MDF can have uncertainty compressed by joint observations in Bayes framework. The reliability framework on MDF was discussed,which has two approaches on integrating the original fusion frameworks.Then a framework based on Multi-criterion Decision Making(MCDM) was proposed that not only defined in decision level but also considered data level. The criterion weights were available by multi-objective optimization, and the simulation results show that the proposed method performs better in anti-risk capability and stability than the mono-criteria and decision-making level method.At last a new data fusion framework was proposed based on Data Field named Multi-source Data Fusion Based on Data Field(DF-MDF). The belief function was built based on potential function which was called Potential Belief Function(PBF), in the condition that the properties of potential function was justified to satisfy data fusion. The model can satisfy the total uncertainty measurement in data level and feature level, which means it can unify various uncertainty in on framework. We had a WSN target location test on DF-MDF and found it more precise and robust than classical methods.