粗糙集和神经网络相结合的数据融合方法研究

【作者】 邢化玲；

【导师】 高社生；

【作者基本信息】 西北工业大学 ， 控制科学与工程， 2007， 硕士

【摘要】 信息科学的飞速发展，对数据处理技术提出了新的、更高的要求。在多传感器系统中，由于传感器的精度、数据的采集成本、系统组成的多种环节以及外部环境等因素的影响，会导致多传感器系统中信息的不完备性。从而，使得传统的数据处理方法，不能满足工程应用对信息处理快速性和高精度的要求，这给数据融合带来困难。因此，对多传感器系统中不完备信息的处理已成为众多学者广泛关注的热点问题。本文将数据挖掘和数据融合技术相结合，用于多传感器系统中不完备信息的处理，所做的主要工作和创新性研究如下： 1．针对数据挖掘与数据融合技术在功能上互补的特点，研究了数据挖掘与数据融合技术相结合的基本概念和原理，并针对数据融合中先验信息难以获取，以及系统中存在大量冗余数据的问题，研究了基于粗糙集理论的数据融合模型。利用粗糙集理论的属性约简算法，剔除冗余信息得到了最简规则，并利用该规则进行数据融合。 2．研究了粗糙集理论中连续数据的离散化方法，提出了一种基于遗传算法的离散化方法。该方法以离散断点的数目最少和离散化后决策系统的一致性最大为优化目标，最大程度地保证了决策系统的一致性。 3．提出了一种直接在不完备信息系统上进行数据挖掘的方法。该方法从实际应用出发，考虑到属性获取的实时性、难易程度和成本等问题，将属性分层后进行递阶处理，求得最接近原始数据的规则。与经过完备化处理后进行挖掘的方法相比，该方法具有较强的实用性和有效性，并提高了知识约简的速度。 4．针对多传感器系统中不完备信息的融合问题，将数据挖掘方法和神经网络技术相结合，提出了一种基于粗糙集——模糊神经网络的融合系统设计方法。给出了系统的设计原理，设计步骤，系统各部分的算法以及网络的构造方法，并推导了网络的学习公式。仿真结果证明了该融合系统的可行性和有效性。更多还原

【Abstract】 With the rapid development of information science, new and higher requirement is presented to data processing technique. Influenced by such factors as the precision of sensors, the cost of data collection, the variety of data mode and the external environment, the information in multi-sensor system is mostly incomplete. As a result, the traditional data processing methods can’t satisfy the demands of high precision and rapid speed of information processing, which brings new challenge to data fusion. Therefore, the processing of incomplete information in multi-sensor system has been a hot spot concerned by many scholars. With the method of combing data mining and data fusion technology, the dissertation researches the processing method of incomplete information.The main work and innovations are as follows:Firstly, according to the complementary functions of data fusion and data mining, the basic concepts and the principle of combining both technologies are investigated. Moreover, in order to deal with the redundant data and the difficulty of obtaining prior knowledge, the fusion model based on rough sets is studied. By applying the reduction method, the redundant information is eliminated and the fusion algorithm is extracted.Secondly, a discretization method based on genetic algorithm is proposed. In this method, the minimum set of cuts and the maximum consistency of decision system are the optimizing goals. Thus the consistency of the system is ensured at the maximal extent.Thirdly, a data mining method based on rough sets theory for incomplete information system is proposed. From the practical application, in consideration of the real-time, difficulty and cost during obtaining the attributes, the attributes are partitioned into complete part and incomplete part. Thus the decision system is presented at two layers. Then the reduction method is hierarchically applied to each layer. This mining method has strong application and fast reduction speed.Finally, the problem of incomplete information in multi-sensor system is investigated. By combining the above mining method and neural network, a design更多还原