Page 73 - 2023年第54卷第10期
P. 73

2022,41(7):47 - 60.
                [24] LM,RENB,WUB,etal.Aparametric3Dgeologicalmodelingmethodconsideringstratigraphicinterfaceto
                       pologyoptimizationandcodingexpertknowledge[J].EngineeringGeology,2021,293:1 - 17.
                [25] 闵恺艺.基于多测点模型的面板堆石坝沉降变形预测研究[D].西安:西安理工大学,2021.
                [26] 赵征,冯事 成,宋 梅 雯,等.基 于 XGBoost的 航 空 器 动 态 滑 行 时 间 预 测 方 法 研 究 [J].航 空 工 程 进 展,
                      2022,13(1):76 - 85.
                [27] 李浩维.基于改进 WOA优化参数的 XGBoost入侵检测模型[D].沈阳:辽宁大学,2022.
                [28] 王涛.非线性权重和柯西变异的蝗虫算法[J].微电子学与计算机,2020,37(5):82 - 86.
                [29] 王龙达,王兴成,刘罡.基于角度惩罚距离的收敛因子非线性递减多目标鲸鱼优化改进算法[J].计算机
                       应用研究,2022,39(5):1395 - 1401.
                [30] ZHANGY,ZHOUX,SHIH PC.ModifiedHarrisHawksoptimizationalgorithm forglobaloptimizationproblems
                       [J].ArabianJournalforScienceandEngineering,2020,45(12):10949 - 10974.
                [31] 周挺,杨军,詹祥 澎,等.一 种 数 据 驱 动 的 暂 态 电 压 稳 定 评 估 方 法 及 其 可 解 释 性 研 究 [J].电 网 技 术,
                      2021,45(11):4416 - 4425.
                [32] 赵晓华,亓航,姚莹,等.基于可解释机器学习框架的快速路立交出口风险预测及致因解析[J].东南大
                       学学报(自然科学版),2022,52(1):152 - 161.
                [33] SHAPLEYLS.Avalueforn - personsgames[J].AnnalsofMathematicsStudies,1953,28(7):307 - 318.
                [34] 陈萌.基于 ICU临床数据的脓毒症早期预测算法及其可解释性研究[D].南京:东南大学,2021.



                          IAO- XGBoostensemblelearningmodelforseepagebehavioranalysisof
                                  earth - rockdam andinterpretationofpredictionresults


                    YUHongling,WANGXiaoling,RENBingyu,ZHENGMingwei,WUGuohua,ZHUKaixuan
                       (StateKeyLaboratoryofHydraulicEngineeringSimulationandSafety,TianjinUniversity,Tianjin 300072,China)


                  Abstract:Inviewoftheproblemsoflowcomputationalefficiencyanddifficultyinreal - timeanalysisofdamseep
                  agebehaviorintheexistingseepagenumericalsimulationmethodsofearth - rockdam,andtheproblemsofpoorin
                  terpretabilityoftheexistingsurrogatemodelbasedonmachinelearningalgorithm ,anIAO- XGBoostensemble
                  learningmodelforseepagebehavioranalysisofearth - rockdamisproposed ,andthepredictedresultsareexplained
                  basedontheSHapleyAdditiveexPlanation(SHAP)theory.Onthebasisofusingmulti - geologicalbodyautomatic
                  modelingmethodandCFDtechnologytocalculateandanalyzetheseepagefieldofthedam,thehyper - parameters
                  suchasn_estimators,max_depthandlearning_rateoftheeXtremeGradientBoosting(XGBoost) ensemble
                  learningalgorithmwereoptimizedbytheImprovedAquilaOptimization(IAO)algorithm.Thenapredictionmodel
                  ofseepagebehaviorindexofdambasedonIAO - XGBoostensemblelearningalgorithmwasestablishedtorevealthe
                  complexnonlinearmappingrelationshipbetweentheinputcharacteristicvariablessuchastheupstream anddown
                  streamwaterlevelandpermeabilitycoefficientsofdamfoundationandthesimulatedvalueofseepagebehaviorin
                  dex.Furthermore ,theIAO- XGBoostensemblelearningalgorithm wascombinedwiththeexplainablemachine
                  learningframeworkSHAPtheorytoexcavatethekeyfeaturesaffectingthepredictionresultsofthedamseepagebe
                  haviorindex ,andexplaintheinfluenceofthecharacteristicvariablesonthepredictionresults.Thecasestudy
                  showsthattheIAO - XGBoostensemblelearningalgorithmhashighpredictionaccuracy.ComparedwithIAO - GB
                  DT ,IAO - RF,IAO - DTandIAO - SVRalgorithms,thepredictionaccuracyofIAO - XGBoostensemblelearning
                  algorithm increasesby0.52%, 11.64%, 37.21% and 25.07%, respectively.Compared with thefeature
                  importanceanalysismethodsofIAO - XGBoost ,IAO - GBDTandIAO - RFalgorithms,SHAPtheoryhasstronger
                  modelinterpretabilityandimprovethereliabilityofthepredictionresults.
                  Keywords:earth - rockdam;seepagebehavioranalysis; XGBoost; interpretability; SHAP theory; improved
                  Aquilaoptimizationalgorithm


                                                                                    (责任编辑:李福田)


                                                                                                   2
                                                                                              —   1 0 9 —
   68   69   70   71   72   73   74   75   76   77   78