面向液氨泄漏应急救援区域的XGBoost预测方法研究  

Research on XGBoost predictionmethod for emergency rescue area ofliquid ammonia leakage

作　　者：王海宁 杨威 黄惟 李海航 WANG Hai-ning;YANG Wei;HUANG Wei;LI Hai-hang(College of Quality and Safety Engineering,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学质量与安全工程学院,杭州310018

出　　处：《安全与环境学报》2023年第5期1482-1489,共8页Journal of Safety and Environment

基　　金：国家自然科学基金青年项目(52106185)。

摘　　要：液氨发生泄漏事故后,随着扩散距离的增加,会对人员和环境造成严重的危害。为便于发生泄漏事故后,快速展开应急救援工作,对液氨泄漏事故应急救援区域预测方法开展研究。通过PHAST(Process Hazard Analysis Software Tool)软件模拟液氨泄漏事故工况,建立基于极端梯度提升(Extreme Gradient Boosting XGBoost)的液氨泄漏应急救援区域预测模型。利用网格搜索结合K折交叉验证进行超参数调优,并与随机森林、决策树模型性能进行对比分析。研究结果显示:以预测ERPG 2标准下液氨泄漏扩散距离为例,XGBoost模型预测性能最佳;与决策树和随机森林相比,XGBoost模型的EMAPE分别减少了4.19个百分点和2.37个百分点,ERMSE分别减少了66.74和2.93;基于优化后XGBoost模型液氨泄漏事故应急救援区域预测模型,预测结果R2为0.9978,ERMSE为50.37,EMAPE为2.61%,基本满足工程实践应用。Chemical companies are prone to liquid ammonia leakage during the production of liquid ammonia.Therefore,the purpose of this paper is to be able to quickly and accurately predict the dispersion distance of liquid ammonia storage tank leaks.This paper takes a chemical enterprise liquid ammonia storage tank area as the object.By referring to the“Method on the determination of emergency planning zone for major accidents of toxic gas leakage”,the emergency rescue area is divided into hot,warm,and cold zones.And the ERPG standard was chosen as the basis for the division of the emergency rescue area.The prediction performance of several mainstream machine learning models is compared and analyzed by simulating a variety of liquid ammonia leakage accident conditions as the original data set by PHAST.And the hyperparameter tuning of XGBoost,random forest and decision tree models using grid search are combined with K fold cross-validation.The predictive performance of the three models is validated and analyzed.Finally,an XGBoostbased prediction model for liquid ammonia leakage accident emergency rescue area is established.The conclusions are shown below:As an example of predicting the dispersion distance of liquid ammonia leaks under the ERPG 2 standard,the XGBoost model has the best prediction performance.There are 4.19 precentage point and 2.37 precentage point reductions in EMAPE and 66.74 and 2.93 reductions in ERMSE for the XGBoost model compared to the decision tree and random forest,respectively.A prediction model for the emergency rescue area of liquid ammonia leakage accidents is established based on XGBoost.Taking the prediction of liquid ammonia dispersion distance under ERPG 2 and ERPG 3 standards in the test set as an example,the R2 value of the prediction result is 0.9978,which indicates that the prediction model fits well.The ERMSE value is 50.37 and the EMAPE is 2.61%,which has high accuracy.The predicted diffusion distance and the actual diffusion distance are very close to meeting the engineering prac

关 键 词：公共安全 应急救援 液氨泄漏 机器学习 极端梯度提升 

分 类 号：X937[环境科学与工程—安全科学]