Extracting Natech Reports from Large Databases:Development of a Semi-Intelligent Natech Identificati

• 作 者: Xiaolong Luo Ana Maria Cruz Dimitrios Tzioutzios
• 机 构: Department of Urban Management,Graduate School of Engineering,Kyoto University,Kyoto 615-8540,Japan
• 出 处: 国际灾害风险科学学报（英文版）
• 发表日期: 2020年6期
• 关键词: Data extraction method Machine learning Natechs Natural hazards NRC database

Natural hazard-triggered technological accidents(Natechs) refer to accidents involving releases of haz-ardous materials (hazmat) triggered by natural hazards.Huge economic losses,as well as human health and envi-ronmental problems are caused by Natechs.In this regard,learning from previous Natechs is critical for risk man-agement.However,due to data scarcity and high uncer-tainty concerning such hazards,it becomes a serious challenge for risk managers to detect Natechs from large databases,such as the National Response Center (NRC)database.As the largest database of hazmat release inci-dents,the NRC database receives hazmat release reports from citizens in the United States.However,callers often have incomplete details about the incidents they are reporting.This results in many records having incomplete information.Consequently,it is quite difficult to identify and extract Natechs accurately and efficiently.In this study,we introduce machine learning theory into the Natech retrieving research,and a Semi-Intelligent Natech Identification Framework (SINIF) is proposed in order to solve the problem.We tested the suitability of two super-vised machine learning algorithms,namely the Long Short-Term Memory (LSTM) and the Convolutional Neural Network (CNN),and selected the former for the develop-ment of the SINIF.According to the results,the SINIF is efficient (a total number of 826,078 records were analyzed)and accurate (the accuracy is over 0.90),while 32,841 Natech reports between 1990 and 2017 were extracted from the NRC database.Furthermore,the majority of those Natech reports (97.85％) were related to meteorological phenomena,with hurricanes (24.41％),heavy rains(19.27％),and storms (18.29％) as the main causes of these reported Natechs.Overall,this study suggests that risk managers can benefit immensely from SINIF in analyzing Natech data from large databases efficiently.