基于数据挖掘的生产事故破坏程度评估系统研究与实现

发布时间：2018-03-11 03:24

本文选题：数据整合　切入点：离群点检测　出处：《中国科学院大学(中国科学院沈阳计算技术研究所)》2017年硕士论文　论文类型：学位论文

【摘要】：随着大数据与数据挖掘等技术的发展,越来越多的信息系统接入了数据挖掘的技术,如语音模拟、图像识别等等。本文提出了在生产事故破坏程度评估系统引入数据挖掘技术,根据外部数据、危险源自身特征以及历史数据训练分类器,对生产事故破坏程度得到比较精确的评估结果。本文主要研究内容有以下几点:1.数据整合。在面对多源异构数据整合并且要求实时性的时候,传统方案中并没有很好的解决实时性需求。本文提出了新的数据整合流程,增加了监控数据、自动更新以及知识公式库模块。2.数据整合过程中异常问题的处理。在众多数据整合方案中,往往出现数据异常问题,导致整合效果减弱。本文提出了基于角度的三阶段异常检测算法,有效地解决异常问题,也避免出现可能被误判的的异常数据。3.基于K邻近算法的改进与应用。在生产事故破坏度评估系统中,为了得到快速高效的分类器,本文提出了基于K邻近算法的改进方案,将样本集进行组织分区,压缩样本集数量,从而减少了与样本集的相似性计算。4.工厂事故破坏度系统的设计与实现。本文给出了该系统核心模块的数据接口设计以及部分源代码实现样例,然后展示了采用数据可视化形式的系统页面效果,最后进行了系统性能测试。
[Abstract]:With the development of big data and data mining technology, more and more information systems are integrated into data mining technology, such as voice simulation, image recognition and so on. Training classifiers based on external data, risk source characteristics and historical data, The main contents of this paper are as follows: 1. Data integration. In the face of multi-source heterogeneous data integration and real-time requirements, The traditional scheme does not solve the real-time requirement well. In this paper, a new data integration process is proposed, and the monitoring data is added. Automatic updating and knowledge formula database module .2. handling of abnormal problems in the process of data integration. In many data integration schemes, there are often abnormal data problems, In this paper, a three-stage anomaly detection algorithm based on angle is proposed to solve the anomaly problem effectively. In order to obtain a fast and efficient classifier, this paper proposes an improved scheme based on K-neighborhood algorithm in order to obtain a fast and efficient classifier in order to avoid the occurrence of abnormal data that may be misjudged. 3. In order to obtain a fast and efficient classifier in the evaluation system of failure degree of production accidents, this paper proposes an improved scheme based on K-neighbor algorithm. The sample set is organized and partitioned, and the number of samples is compressed. Therefore, the similarity calculation with the sample set is reduced. 4. The design and implementation of the factory accident damage degree system. This paper presents the data interface design of the core module of the system and some examples of the source code implementation. Then the system page effect of data visualization is demonstrated, and the system performance test is carried out at last.
【学位授予单位】：中国科学院大学(中国科学院沈阳计算技术研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.52;TP311.13

【参考文献】