Multi-Mode Intelligent Reasoning And Fusion Based On Generalized Evidence Theory

When all kinds of information fusion technology tend to be mature, a unified approach or framework that can combine all kinds of reasoning has not appeared. Due to the fact that the existing reasoning and fusion algorithms diversify, single reasoning module has shortcomings, weak versatility, strong limitations, strong dependence of background and status, this paper researched on how to build a versatile reasoning and fusion framework and tool that was able to combine the advantages of each method of reasoning and could improve the recognition rate of the results of reasoning and fusion. The following aspects were discussed:(1) The four categories of uncertain information, random uncertain information, fuzzy uncertain information, gray uncertain information and unascertained uncertain information were summarized. The four categories of uncertain information were corresponded to three kinds of forms, the probability, subordinate degree function and the interval. Methods of how to turn these four types of uncertainty information to the corresponding basic probability assignment and generalized basic probability assignment were summarized.(2) On the basis of the determination of basic probability assignment based on interval numbers, and combine the generalized evidence theory in the open world, the paper proposed an approach to determine generalized basic probability assignment based on the interval number, which offered a new idea of the determination of generalized basic probability besides the determination based on fuzzy theory. The rationality and effectiveness were verified by the experiments.(3) The framework of multi-mode intelligent reasoning and fusion was proposed. It used the module that Generalized Evidence Theory inference engine, Bayesian Network inference engine, Neural Network inference engine inferred parallel, and fused the inference results by Generalized Evidence Theory.(4) The software of multi-mode intelligent reasoning and fusion engine was developed on C++ and Qt platform. A verification experiment was carried out in the background of Airborne Missile Operational Effectiveness Evaluation. The analysis of multiple sets of comparative test experiment showed the advantage that when one of the inference engine had deviation, multi-mode intelligent reasoning and fusion would neglect it. The experiment showed the stability and effectiveness of the algorithm.