三种机器学习分类在LNG泄漏风险评估中的比较
|
摘要
根据LNG储运行业法律法规要求,结合企业真实生产情况和相关专家意见,梳理出LNG泄漏风险评估体系,包括5个一阶评估指标和20个二阶评估指标。对多家真实企业LNG泄漏风险管理及相关后果进行评估,得出37条样本组成的数据集。将该数据集运用核支持向量机、随机森林以及神经网络三种分类模型处理,比较三种机器学习分类模型在分类准确率上的差别;根据分类效果研究相关参数对准确率的影响。结果认为,核支持向量机和神经网络需要进行细致的参数调整以保证较好的准确率,随机森林受到参数影响效果较小。根据分类指标权重的比较提出相应的控制风险对策措施。
This research first summarized the system of LNG leakage risk, according to the requirements of the laws and regulations in LNG storage and transportation industry, and combined with the situation of enterprises and opinions of experts. The system included 5 first-order indicators and 20 second-order indicators.This research obtained a data set of 37 samples, by assessing the risk management and consequences in real enterprises. This data set was applied to three classification models, namely SVM, Random Forest Classification and Neural Network, to compare the differences of accuracy among these machine learning classification models, and to study the influence of relevant parameters on accuracy. It was concluded that SVM and Neural Network needed to adjust the parameters carefully to ensure a better accuracies, while the Random Forest Classification was not affected by the parameters. Corresponding risk control measures were put forward by compared the weights of second-order indicators.
This research first summarized the system of LNG leakage risk, according to the requirements of the laws and regulations in LNG storage and transportation industry, and combined with the situation of enterprises and opinions of experts. The system included 5 first-order indicators and 20 second-order indicators.This research obtained a data set of 37 samples, by assessing the risk management and consequences in real enterprises. This data set was applied to three classification models, namely SVM, Random Forest Classification and Neural Network, to compare the differences of accuracy among these machine learning classification models, and to study the influence of relevant parameters on accuracy. It was concluded that SVM and Neural Network needed to adjust the parameters carefully to ensure a better accuracies, while the Random Forest Classification was not affected by the parameters. Corresponding risk control measures were put forward by compared the weights of second-order indicators.
引文
[1]何世明.基于机器学习及群智能寻优算法的化工故障诊断建模优化研究[D].杭州:浙江大学,2018.
[2]郑帅.化工突发事件信息抽取方法的研究[D].青岛:青岛科技大学,2017.
[3]冯晓秀.基于支持向量机的宁东能源化工基地环境问题研究[D].银川:宁夏大学,2015.
[4]关文玲,王少莉,朱晓莉.基于支持向量机的化工企业安全预警模型研究[J].天津理工大学学报,2017,33(4):16-20.
[5]徐绪堪,王京.基于随机森林的突发事件分级模型研究[J].中国安全生产科学技术,2018,14(2):77-81.
[6]赖成光,陈晓宏,赵仕威,等.基于随机森林的洪灾风险评价模型及其应用[J].水利学报,2015,46(1):58-66.
[7]王中锋.基于动态朴素贝叶斯分类器的明渠水华风险评估模型[J].南水北调与水利科技,2017,15(2):89-94.
[8]单彤文. LNG储罐研究进展及未来发展趋势[J].中国海上油气,2018,30(2):145-151.
[9]李航.统计学习方法[M].北京:清华大学出版社,2012.
[10] MüLLER C A,GUIDO S. Introduction to machine learning with Python[M]. Post&Telecom Press,2018.
[11] PLAS V J. Python data science handbook[M]. Post&Telecom Press,2018.
[12]张雪蕾,汪明,曹寅雪,等. 3种典型机器学习方法在灾害敏感性评估中的对比[J].中国安全生产科学技术,2018,14(7):80-85.
[13]谭蔚,王康,薛同辉,等. LPG储配站介质泄漏扩散过程模拟[J].消防科学与技术,2018,37(3):296-299.
[14]王丽.接收站泄漏LNG气化及扩散规律研究[J].消防科学与技术,2016,35(12):1762-1766.
[15]安春晖.某CNG加气站的火灾爆炸危险性评价[J].消防科学与技术,2014,33(7):828-832.
[2]郑帅.化工突发事件信息抽取方法的研究[D].青岛:青岛科技大学,2017.
[3]冯晓秀.基于支持向量机的宁东能源化工基地环境问题研究[D].银川:宁夏大学,2015.
[4]关文玲,王少莉,朱晓莉.基于支持向量机的化工企业安全预警模型研究[J].天津理工大学学报,2017,33(4):16-20.
[5]徐绪堪,王京.基于随机森林的突发事件分级模型研究[J].中国安全生产科学技术,2018,14(2):77-81.
[6]赖成光,陈晓宏,赵仕威,等.基于随机森林的洪灾风险评价模型及其应用[J].水利学报,2015,46(1):58-66.
[7]王中锋.基于动态朴素贝叶斯分类器的明渠水华风险评估模型[J].南水北调与水利科技,2017,15(2):89-94.
[8]单彤文. LNG储罐研究进展及未来发展趋势[J].中国海上油气,2018,30(2):145-151.
[9]李航.统计学习方法[M].北京:清华大学出版社,2012.
[10] MüLLER C A,GUIDO S. Introduction to machine learning with Python[M]. Post&Telecom Press,2018.
[11] PLAS V J. Python data science handbook[M]. Post&Telecom Press,2018.
[12]张雪蕾,汪明,曹寅雪,等. 3种典型机器学习方法在灾害敏感性评估中的对比[J].中国安全生产科学技术,2018,14(7):80-85.
[13]谭蔚,王康,薛同辉,等. LPG储配站介质泄漏扩散过程模拟[J].消防科学与技术,2018,37(3):296-299.
[14]王丽.接收站泄漏LNG气化及扩散规律研究[J].消防科学与技术,2016,35(12):1762-1766.
[15]安春晖.某CNG加气站的火灾爆炸危险性评价[J].消防科学与技术,2014,33(7):828-832.