面向用户偏好的突发事件知识获取研究

发布时间：2018-05-21 01:08

本文选题：知识获取 + 突发事件　；参考：《大连理工大学》2016年硕士论文

【摘要】：随着我国对突发事件重视程度与信息化程度的提高,一些相关数据库中积累了各类突发事件的大量原始数据,其中蕴含着许多潜在、有价值的突发事件规律知识。而数据收集过程中由于自然因素和人为因素导致的随机性、盲目性,容易造成信息冗余问题,阻碍潜在知识规律的发现。现有知识获取研究大多以知识的客观性与简洁性为目标,忽略了用户对知识规则的偏好,其知识获取结果实用性较差,不利于辅助突发事件应急决策。因此,本文在粗糙集理论的基础上,提出一种面向用户偏好的突发事件知识获取方法,主要包括预处理、属性约简与规则提取三部分。首先,对原始信息表进行知识获取预处理操作。对突发事件及其数据特征进行分析,将应急决策者对知识获取的偏好总结为属性选择偏好、属性粒度偏好与规则信度偏好三种,并对用户偏好进行定量表示；识别当前决策属性,构建决策表；针对突发事件数据特征,提出缺失数据、冗余数据、噪声数据以及多类型多粒度数据的预处理方法,通过构建决策表相似性矩阵识别并剔除冗余、噪声数据,利用属性概念层次树层次化表示多类型多粒度数据；为了提高决策表处理效率,编码决策表。其次,结合突发事件数据特征与用户偏好,提出一种突发事件属性约简方法。该方法综合属性客观重要度和用户属性选择偏好两方面的影响定义启发信息,利用属性添加的启发式方法求得一个约简结果。其中为了简化属性重要度的计算,提出一种基于OWA算子的度量方法,综合属性在决策表行和列两个维度的贡献计算属性对各单个对象的重要度,利用OWA算子将其聚合为属性的综合重要度。最后,针对突发事件知识获取中知识粒度难以确定、知识规则实用性差的问题,提出一种面向用户偏好的层次化规则提取方法。该方法从最高层次的广义决策表开始,以属性客观分类能力与用户属性粒度偏好作为衡量标准选择属性进行特化,使决策表层次下降；针对下降过程中得到的每一个特定层次的广义决策表,对人工蜂群算法进行改进以便高效的提取其中的确定性规则；以满足用户规则信度偏好作为层次下降的终止条件,对知识粒度加以控制。本文在粗糙集理论的基础上对突发事件数据进行知识获取,该方法引入用户偏好及层次规则提取的思想,解决了信息缺失、属性类型多样、知识粒度难以确定、规则实用性差等问题,在满足用户对于知识规则实际应用需求的前提下,获取较为简洁的知识,能够辅助应急管理决策,达到防灾、减灾的目的,具有一定的理论与实际意义。
[Abstract]:With the improvement of the degree of attention and information of unexpected events in China, a large number of raw data of all kinds of emergencies have been accumulated in some related databases, which contains a lot of potential and valuable knowledge of the law of emergencies. The randomness and blindness caused by natural and human factors in the process of data collection can easily lead to the problem of information redundancy and hinder the discovery of the law of potential knowledge. Most of the existing knowledge acquisition studies aim at the objectivity and conciseness of knowledge, ignoring the user's preference for knowledge rules, and the practical results of knowledge acquisition are poor, which is not conducive to assisting emergency decision-making. Therefore, based on rough set theory, this paper proposes a knowledge acquisition method based on user preference, which consists of three parts: preprocessing, attribute reduction and rule extraction. First of all, the original information table is preprocessed with knowledge acquisition. The emergency decision makers' preference for knowledge acquisition is summarized as attribute preference, attribute granularity preference and rule reliability preference. Identify the current decision attributes, construct decision table, propose a preprocessing method of missing data, redundant data, noise data and multi-type and multi-granularity data according to the characteristics of emergency data. The similarity matrix of decision table is constructed to identify and eliminate redundant and noise data, and hierarchical attribute concept tree is used to represent multi-type and multi-granularity data. In order to improve the efficiency of decision table processing, the decision table is coded. Secondly, an attribute reduction method is proposed based on the feature of emergency data and user preference. The method synthesizes the heuristic information of attribute objective importance and user attribute selection preference, and obtains a reduction result by using the heuristic method of attribute addition. In order to simplify the calculation of attribute importance, a measurement method based on OWA operator is proposed, which synthesizes the contribution of attributes in the decision table row and column dimensions to calculate the importance of attributes to each single object. The OWA operator is used to aggregate it into the synthetic importance of attributes. Finally, a hierarchical rule extraction method based on user preference is proposed to solve the problem that the granularity of knowledge is difficult to determine and the practicability of knowledge rules is poor. This method starts with the highest level of generalized decision table and uses attribute objective classification ability and user's attribute granularity preference as the standard to select attributes. For each generalized decision table at a specific level in the descent process, the artificial bee colony algorithm is improved to extract deterministic rules efficiently, and to satisfy the user rule reliability preference as the termination condition for hierarchical descent. Control the granularity of knowledge. On the basis of rough set theory, this paper introduces the idea of user preference and hierarchical rule extraction into the knowledge acquisition of unexpected event data, which solves the problem of missing information, diverse attribute types, and difficult to determine the granularity of knowledge. Problems such as poor practicability of rules, on the premise of satisfying the practical application needs of users for knowledge rules, can obtain more concise knowledge, which can assist emergency management decisions and achieve the purpose of disaster prevention and mitigation, which has certain theoretical and practical significance.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP18

【相似文献】