非完备信息的基本概率指派生成及应用研究

Due to the rapid development of science and technology and the tough environment of the battlefield, the modern fighter multi-sensor target recognition system has to process a large number of incomplete uncertain information. The conventional DS evidence theory is based on complete proposition, and it can neither model the incomplete information nor get a quantitative description of it. What’s more, it can’t solve the conflict combination problem caused by incomplete frame of identification. Thus the performance of the fusion system is greatly affected. In order to address these issues, based on the proposed generalized evidence theory, this project breaks the frame of a closed world in the conventional evidence theory to explore the generation of open world basic probability assignment (BPA), and to represent the various heterogeneous incomplete uncertain information with different structures, as well as to establish a model to determine whether the frame of identification of a system is complete. If the frame of discernment is incomplete, the model database is updated. And this will solve the problem of conflict evidence combination fundamentally in the case of incomplete frame of identification. As to decision making, this project explores the relationship between the BPA and the probability, after that a new evaluation criterion of probability transition and a mathematical model of the optimal probability distribution for the BPA transformation are presented for the final decision of a system. The research has great significance to the information fusion research based on evidence theory.

科学技术的高速发展以及战场环境的恶劣，使得现代机载多传感器目标识别系统不得不面临大量非完备不确定信息的处理。经典DS证据理论建立在命题完备基础上，无法实现对非完备信息的建模和定量描述，且不能解决由于辨识框架不完备所引起的冲突融合问题，极大影响了机载信息融合系统性能。针对这些问题，本项目突破经典证据理论的封闭世界框架，基于所提出的广义证据理论，研究开放世界下的非完备信息基本概率指派（Basic Probability Assignment, BPA）生成，实现对各种异类异构非完备不确定信息的表示；建立判断系统辨识框架是否完备的模型；针对辨识框架不完备情况，实现目标模型库的更新；从根本上解决辨识框架不完备情况下的冲突证据融合；在决策阶段，探索BPA和概率之间的关联关系，提出新的概率转换评价准则和最优BPA转换概率分布的数学模型用于系统最终决策。本项目对基于证据理论的信息融合研究具有重要意义。

科学技术的高速发展以及战场环境的恶劣，使得现代机载多传感器目标识别系统不得不面临大量非完备不确定信息的处理。经典DS证据理论建立在命题完备基础上，无法实现对非完备信息的建模和定量描述，且不能解决由于辨识框架不完备所引起的冲突融合问题，极大影响了机载信息融合系统性能。针对这些问题，本项目突破经典证据理论的封闭世界框架，基于所提出的广义证据理论，研究开放世界下的非完备信息基本概率指派（Basic Probability Assignment, BPA）生成，提出了基于样本差异度的非完备信息BPA生成方法，实现对各种非完备不确定信息的表示，并基于证据关联系数建立了参数冲突表示模型；针对辨识框架不完备情况，提出了多维度目标库完备性判别方法以及基于高斯重叠度的目标库完备性判别方法，有效实现了系统辨识框架完备性的判别；并提出了基于聚类的目标模型库更新方法，将未知目标类型的样本更新为不同的未知类别，完善模型库。最后，在决策阶段，探索BPA和概率之间的关联关系，提出了基于证据关联系数的信度决策方法，有效解决证据理论应用时的决策问题。本项目对非完备条件下基于证据理论的信息融合研究具有重要意义。

内容获取失败，请点击重试