核主元分析与优化核极限学习机模型在电石炉爆炸风险评估中的应用  

Application of Kernel Principal Component Analysis and optimized Kernel Extreme Learning Machine model in risk assessment of calcium carbide furnace explosion

作　　者：毕颖 马世杰 BI Ying;MA Shijie(College of Environmental and Safety Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学环境与安全工程学院,沈阳110142

出　　处：《安全与环境学报》2024年第6期2075-2084,共10页Journal of Safety and Environment

摘　　要：为准确判断电热法电石生产工艺中电石炉的爆炸风险等级,提出了一种精准有效的风险评估模型。首先,基于危险与可操作性(Hazard and Operability, HAZOP)分析筛选出人、物料、设备、管理四方面的34项爆炸风险因素,考虑到因素间存在非线性关联,采用核主元分析(Kernel Principal Component Analysis, KPCA)进行属性约简,减少冗杂信息的干扰。其次,利用融合了Tent混沌序列、高斯变异与混沌扰动的麻雀搜索算法(Improved Sparrow Search Algorithm, ISSA)寻优核极限学习机(Kernel Extreme Learning Machine, KELM)的惩罚系数与核参数,建立KPCA-ISSA-KELM风险评估模型。最后,使用该模型分析83组实例数据,选取其中59组用于模型训练,其余24组用于测试。在测试结果中,该模型正确分类了22组数据的风险等级,判别准确率为91.67%,在各项性能指标上均优于对照模型,表明该模型对电热法工艺电石炉的爆炸风险等级具备高识别精度。Electrothermal calcium carbide production is a complex process involving multiple factors.As the main production equipment,sealing-type calcium carbide furnace carries a relatively high explosion risk that is difficult to quantify.To accurately determine the explosion risk level of calcium carbide furnaces and prevent related accidents,this paper proposed an accurate and effective risk assessment model.First,based on Hazard and Operability(HAZOP)analysis and accident-causing theories,34 risk factors related to calcium carbide furnace explosion covering personnel,materials,equipment,and management were selected.Due to the complex correlation between factors,this paper introduced Kernel Principal Component Analysis(KPCA)to perform nonlinear feature extraction and dimensionality reduction on the factors,achieving reconstruction of the factor system and reducing interference from redundant information.After that,incorporating tent chaotic sequence,gaussian mutation,and chaotic perturbation,the Sparrow Search Algorithm(SSA)was improved.The result of the benchmark function optimization test shows that:the Improved Sparrow Search Algorithm(ISSA)is superior to SSA and Particle Swarm Optimization(PSO)in terms of optimization accuracy and convergence speed.Subsequently,ISSA was used to optimize the penalty coefficient C and kernel parameter s of the Kernel Extreme Learning Machine(KELM),and the KPCA-ISSA-KELM risk assessment model was established.83 sets of original data were obtained from several enterprises,including safety inspection results and corresponding discrimination from experts on the risk level of calcium carbide furnace explosion.Among them,59 sets were used for training the model,and the remaining 24 sets were used for testing.Programming of models was carried out having the aid of MATLAB software.The nonlinear feature extraction result shows that KPCA obtained 11 principal components that contained 85.86%information of the original data.Moreover,the risk prediction result of 24 test set samples shows that:K

关 键 词：安全工程 风险评估 电石炉 核主元分析(KPCA) 麻雀搜索算法(SSA) 核极限学习机(KELM) 

分 类 号：X932[环境科学与工程—安全科学]