【作者】 胡学海；

【导师】 王厚军；

【作者基本信息】 电子科技大学 ， 测试计量技术及仪器， 2008， 博士

【摘要】 随着战场环境日益复杂恶劣,为提高飞机武器系统的作战性能,研究复杂干扰环境下机载多传感器目标识别问题具有极其重要的意义。为此,本文研究了复杂干扰条件下的机载多传感器属性融合问题,主要取得的成果为:1)本文首先分析了复杂干扰环境下属性融合存在融合信息不确定性大,造成系统鲁棒性和自适应性变差的问题,提出一种自适应融合模型。模型分三个部分:不确定信息处理、分类识别信息融合、知识更新。其中知识更新模块利用智能技术解决不同环境和信息层次下融合系统所要求的环境自适应性,稳健性问题。在此基础上,本文研究了知识更新模块的建立,提出利用蚁群算法进行知识的智能分析和推理寻优,从而实现在复杂环境下知识和更新的自适应管理。为进一步提高算法性能,本文研究了蚁群算法的改进问题,提出了爬山变异蚁群算法,将蚂蚁算法和局部搜索算法结合起来,提高算法后期的计算效率。2)针对数据层属性融合问题,本文分别研究了T.Fukuda和Luo R.C提出的不同算法,在此基础上,本文提出了一种基于传感器信息的一致性和可信度的分组融合算法。算法通过修正类别置信度的分类估计,减小测量误差引起的目标分类置信度的估计偏差;通过一致性测度进行传感器分组,利用传感器先验可信度估计各分组传感器的可信度,再优选可信度高的传感器组进行融合,从而修正冲突信息对融合结果的影响。该算法将不确定信息处理和融合结构有机结合,提高了属性融合的抗干扰能力。3)针对硬决策属性融合问题,本文首先研究了目标先验知识已知的情况下,分布式检测系统的算法优化算法,如穷举搜索算法、混合搜索算法,发现这些算法大多是次优算法,所以本文提出了基于爬山变异蚁群算法的贝叶斯全局优化算法。然后针对复杂环境及目标先验知识未知情况,研究了主观BAYES算法、NP算法,发现这些算法的性能和自适应性很难同时提高,所以本文提出了基于一致性分组和分类学习策略和爬山变异蚁群算法的贝叶斯全局优化算法,该算法具有全局优化能力和环境的自适应性。4)针对软决策属性融合问题,本文在深入研究模糊积分的基础上,改进传感器的模糊积分密度的定义,提出一种采用爬山变异蚁群算法确定统计模糊积分密度,采用环境模糊积分密度修正环境干扰的学习算法,并对比了采用遗传算法、神经网络等不同算法的性能,发现本文算法能明显提高算法的环境自适应性和在复杂环境下的融合性能较优。5)最后,本文研究了在不同的机动条件及电磁干扰条件下,上述算法在机载平台战场环境下的目标识别和机载流量传感器故障识别中的应用,验证了算法的有效性。更多还原

【Abstract】 With the battlefield getting increasingly complex, it is significant to study airborne multi-sensor target identification in a complex interference environment in order to improve performance of aircraft weapon system. So, in this paper, the attribute fusion problem of airborne multi-sensor is mainly studied. The results are as following:1) Firstly, based on an analysis of the problems of attribute fusion in a complex interference environment, such as uncertainty of fusion information to make the system’s robustness and self-adaptation worse, a self-adaptive fusion model is brought forward. The model is constituted by three parts: uncertain information processing, classified information fusing and knowledge updating. The knowledge updating utilizes intelligence technology to meet the requirements of environmental self-adaptation and robustness of data fusion system working in a changeful environment and information level. Based on that model, this paper studies foundation of information-updating module and then prensents knowledge’s intellectualized analysis and consequence optimization by ant colony algorithm to realize self-adaptation management of knowledge and updating in a complex environment. To improving algorithm, the paper researches amelioration problem of ant colony algorithm, and then presents Ant System with Bit Climbing Aberrance, which combine ant colony algorithm with part-search to improve arithmetic’s evening calculative efficiency.2) In view of attribute fusion problems in data level, this paper researches different algorithm presented by T.Fukuda and Luo R.C and then presents grouping fusion algorithm based on consistency and reliability of sensors’ information. This algorithm amends classification estimation of the classification confidence to reduce the estimation error of object’s classification confidence made by measure error; classes sensors by consistency measure and estimates the reliability of different sensor groups by the sensors’ transcendent reliability; lastly, chooses the best sensor group of reliability to fuse, which can amend influence of conflict information. This algorithm combines uncertain information processing with the fusion structure organically, which enhances anti-interference capability of attribute fusion.3) In view of attribute fusion problem of hard-decision, this paper studies algorithm optimization of the distributed detection system when object transcendent information was known, such as searching algorithm and mixed searching algorithm. After discovering that these algorithms are inferior-optimizing, this paper presents Bayesian global-optimization algorithm based on Ant System with Bit Climbing Aberrance. For the complex environment and unknown background knowledge about targets, this paper discovers that subjective BAYES and NP algorithm is difficultly to improve the performance and self-adaptation at the same time, and then brings forward BAYES global-optimization algorithm based on consistency grouping, strategy of classified learning and Ant System with Bit Climbing Aberrance. This algorithm has self-adaptation of environment and is global-optimizing.4) In view of attribute fusion problem of soft-decision, on the base of fuzzy integral, this paper gives a new definition about density of fuzzy integral, presents an algorithm which makes use of Ant System with Bit Climbing Aberrance to ascertain Statistic integral density, uses Environment integral density to amend environment molestation and discovers that this algorithm can enhance self-adaptation of environment and performance of this algorithm is better, after contrasting performance of genetic algorithm and NN algorithm.5) Lastly, this paper applies the above algorithm to target identification in airborne battlefield environment and failure diagnosis of airborne flux- temperature sensors in the different condition of aviation and electromagnetism interference, which validates its effectiveness.更多还原