Hazardous Chemicals Accident Prediction Based on Accident State Vector Using Multimodal Data

被引:0
作者
Liu, Kang-Wei [1 ,2 ]
Wan, Jian-Hua [1 ]
Han, Zhong-Zhi [1 ]
机构
[1] China Univ Petr, Sch Geosci, Qingdao 266580, Peoples R China
[2] Sinopec Safety Engn Inst, 339 Songling Rd, Qingdao 266071, Shandong, Peoples R China
来源
FUZZY SYSTEMS AND DATA MINING II | 2016年 / 293卷
关键词
hazardous chemical accidents; Support Vector Machine; accident prediction; accident state vector;
D O I
10.3233/978-1-61499-722-1-232
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Hazardous chemicals industry is a high-risk industry, all kinds of explosions, fires, leaks and poisoning incidents is occurred occasio(1)nally. So it is particularly important to forecast for hazardous chemical accidents and develop appropriate safety measures. In view of the analysis and summary of previous methods, an improved Hazardous Chemicals Accident Prediction method is proposed based on accident state vector in this paper. It defines the accident state vector using Multi-modal Data such as authoritative data, accident report, webpage, image, video, speech, etc. The Multi-modal Data is collected by web crawler which is built by open-source tools. The web crawler is an Internet bot which systematically browses the known hazardous chemical accident website, for the purpose of collecting Multi-modal accident data. As mentioned before, the Multi-modal Data is Multi format. In order to define the accident state vector easily, we divide the Multi-modal data into three dimensions based on the principle of accident causes. Respectively is the human factors, physical state factors, environmental factors. According to the geometrical distribution characteristics of support vector, it can be selected from the incremental samples that the sample of support vectors most likely to become forming a boundary vector set by adopting vector distance pre-extraction method, on which support vector training and accident prediction model build. It ensures the validity of predictive models due to various factors of the cause of the accident are fully considered by the accident state vector and advantages of support vector machines in high-dimensional, multi-factor, large sample datasets machine learning are exhibited. Sample experimental verification from the mastered accident of hazardous chemicals has showed that hazardous chemical accident prediction method proposed in this paper can effectively accumulate accident history information, possess higher learning speed and be positive significance for the safe development of hazardous chemicals industry.
引用
收藏
页码:232 / 240
页数:9
相关论文
共 17 条
  • [1] [Anonymous], 1999, INT JOINT C ART INT
  • [2] Cauwenberghs G., 2000, ADV NEURAL INFORM PR, P122
  • [3] LIBSVM: A Library for Support Vector Machines
    Chang, Chih-Chung
    Lin, Chih-Jen
    [J]. ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY, 2011, 2 (03)
  • [4] Cristianini N., 2004, INTRO SUPPORT VECTOR, P543
  • [5] An efficient support vector machine cone programming for learning method with second-order large-scale problems
    Debnath, R
    Muramatsu, M
    Takahashi, H
    [J]. APPLIED INTELLIGENCE, 2005, 23 (03) : 219 - 239
  • [6] Debnath R, 2004, 2004 2ND INTERNATIONAL IEEE CONFERENCE INTELLIGENT SYSTEMS, VOLS 1 AND 2, PROCEEDINGS, P520
  • [7] Incremental support vector machine construction
    Domeniconi, C
    Gunopulos, D
    [J]. 2001 IEEE INTERNATIONAL CONFERENCE ON DATA MINING, PROCEEDINGS, 2001, : 589 - 592
  • [8] Heaton J., 2002, NET ROBOT JAVA PROGR, V22, P1
  • [9] A simple decomposition method for support vector machines
    Hsu, CW
    Lin, CJ
    [J]. MACHINE LEARNING, 2002, 46 (1-3) : 291 - 314
  • [10] Katagiri S, 2005, IEEE SYS MAN CYBERN, P1258