油气场站火灾爆炸风险的神经支持决策树识别与预测  

Risk recognition and prediction of fire and explosion in oil and gas stations based on neural-backed decision trees

作　　者：闵超[1,2,3] 张乾 黄鑫 龙梦舒 李柯江 刘凤珠 MIN Chao;ZHANG Qian;HUANG Xin;LONG Mengshu;LI Kejiang;LIU Fengzhu(School of Science,Southwest Petroleum University,Chengdu 610500,China;Institute for Artificial Intelligence,Southwest Petroleum University,Chengdu 610500,China;National Key Laboratory of Oil and Gas Reservoir Geology and Exploitation,Southwest Petroleum University,Chengdu 610500,China;Institute for Safety,Environmental Protection and Technical Supervision,PetroChina Southwest Oil and Gas Field Branch,Chengdu 610000,China)

机构地区：[1]西南石油大学理学院,成都610500 [2]西南石油大学人工智能研究院,成都610500 [3]西南石油大学油气藏地质及开发工程全国重点实验室,成都610500 [4]中国石油西南油气田公司安全环保与技术监督研究院,成都610000

出　　处：《安全与环境学报》2024年第7期2574-2584,共11页Journal of Safety and Environment

基　　金：四川省科技创新苗子工程项目(2022034)。

摘　　要：为了有效防控油气场站火灾爆炸事故,从影响因素之间因果关系的角度出发,提出利用神经支持决策树(Neural-Backed Decision Tree,NBDT)算法构建油气火灾爆炸可解释预测模型。该方法利用词频逆向文件频率(Term Frequency-Inverse Document Frequency,TF-IDF)算法从风险描述信息中提取出关键词并计算权重,整合得到64个风险二级因素,构建了油气场站的火灾爆炸数据集;采用神经支持决策树算法构建分类模型,对油气场站火灾爆炸事故进行预测和可解释分析,可以基于数据可视化地分析油气火灾爆炸事故的风险与诱因。结果表明,NBDT模型预测准确率为0.976,AUC为0.913,明显优于其他模型;模型可视化结果分别从单因素和多因素角度分析,确立7种二级风险主控因素和6种二级风险组合主控因素。13种风险主控因素的确立,可以为既有油气场站火灾爆炸预测和防控机制提供理论支撑。To effectively prevent and control fire and explosion incidents at oil and gas facilities,this study proposes utilizing the Neural-Backed Decision Tree algorithm(NBDT)to construct an interpretable predictive model.This approach considers the causal relationships among influencing factors,offering insights into the occurrence of fire and explosion incidents at these facilities.The model facilitates visual analysis of the risks and inducements associated with such incidents by leveraging data.Initially,this method utilizes tokenization techniques and Term Frequency-Inverse Document Frequency(TF IDF)to perform text analysis on 13 primary risk factors,leading to the identification of 64 secondary risk factors.These 64 secondary risk factors are then binary-encoded to construct a dataset for fire and explosion incidents at oil and gas facilities.Subsequently,the study employs the NBDT algorithm to construct a classification model and further optimizes it.The established model is subsequently utilized to predict and interpret fire and explosion incidents at oil and gas facilities.Based on actual data from fire and explosion incidents at oil and gas facilities,this study conducts comparative experimental analyses,comparing the NBDT method against Artificial Neural Networks(ANN)and CART decision tree models.Subsequently,the study utilizes the NBDT model training results to create a visual decision tree diagram.Additionally,single-factor and multi-factor interpretative analyses are conducted based on the mechanisms of fire and explosion incidents,analyzing the causes,and proposing corresponding measures.The experimental results indicate that,compared to ANN and CART decision trees,the NBDT model exhibits higher accuracy of 0.976,precision of 0.985,recall of 0.971,and AUC of 0.913 values on both the training and testing sets.Through single-factor analysis,the NBDT decision tree identifies seven primary control factors for secondary risks.Through multi-factor analysis,seven combinations of secondary risk factors are identif

关 键 词：安全工程 油气爆炸 风险因素 关联规则 可解释性 神经支持决策树(NBDT) 

分 类 号：X928.7[环境科学与工程—安全科学]