Page 119 - 2025年第56卷第8期
P. 119
[32] 刘戎.螺旋隧道施工区域污染扩散机制及控制技术研究[D].重庆:重庆大学,2020.
[33] 中国水利水电第十四工程局有限公司.水工建筑物地下工程开挖施工技术规范:DL?T5099—2011[S].北
京:中国水利水电出版社,2011.
[34] 王晓玲,李清梦,刘宗显,等.融合时频空间特征的土石坝地震易损性分析改进 MLP模型研究[J].水利
学报,2024,55(1):13 - 23.
[35] 王晓玲,李克,张宗亮,等.耦合 ALO - LSTM 和特征注意力机制的土石坝渗压预测模型 [J].水利学报,
2022,53(4):403 - 412.
[36] 班多晗,吕鑫,王鑫元.基于一维混沌映射的高效图像加密算法[J].计算机科学,2020,47(4):278 - 284.
[37] 何庆,罗仕杭.混合改进策略的黑猩猩优化算法及其机械应用[J].控制与决策,2023,38(2):354 - 364.
[38] JASK,DODAGOUDARGR.Explainablemachinelearningmodelforliquefactionpotentialassessmentofsoilsu
singXGBoost - SHAP [J].SoilDynamicsandEarthquakeEngineering,2023,165:107662.
IHPO- XDFrobustpredictionmodelforventilationfrequencyof
undergroundpowerhousecaverns
1
1
1
2
1
1
WANGXiaoling,GUOZhangchao,YUJia,YUHongling,LIUChangxin,WUBinping
(1.StateKeyLaboratoryofHydraulicEngineeringIntelligentConstructionandOperation,TianjinUniversity,Tianjin 300072,China;
2.CollegeofWaterResourcesandCivilEngineering,ChinaAgriculturalUniversity,Beijing 100083,China)
Abstract:Intheconstructionofhydropowerundergroundpowerplantcaverncomplex,thearrangementofthecav
ernsiscomplex ,withmanydeadendsanddisorderedairflowpatterns.Determininganappropriateconstruction
ventilationfrequencyiscrucialforensuringsafety.Duringtheconstructionprocessoftheundergroundpowerplant
caverncomplex ,electromagneticinterferenceandblastvibrationoftenresultinnoisyandmissingdatainenviron
mentalmonitoring ,andexistingmachinelearning - basedconstructionventilationfrequencypredictionmodelsare
highlysensitivetooutliers ,withpoorrobustness.Toaddresstheseissues,thispaperproposesanIHPO - XDFro
bustpredictionmodelforconstructionventilationfrequencyinundergroundpowerplantcaverncomplexes.First ,a
deepforest (DF)networkwithstronguniversallearningabilityandrobustnessisselectedasthebasisforventilation
frequencyprediction.Second ,therandom forestbaselearnerinthedeepforestisimprovedtoextremegradient
boostingtree (XGBoost),andtheefficientgradientboostingmechanismandregularizationstrategyofXGboostare
usedtofurtherenhancetherobustnessandgeneralizationabilityofthemodel.Furthermore ,theImprovedHunter -
PreyOptimization(IHPO)algorithmwasadoptedtooptimizethehyperparametersoftheDFmodel,whichcould
compensatefortheshortcomingsoftraditionalmanualparametertuningandobtaintheoptimalhyperparameters,
therebyminingandenhancingtheinherentrobustnessofthemodelandrevealingthecomplexnonlinearmappingre
lationshipsbetweenthepollutantconcentration ,temperature,windspeed,andexcavationvolumeandthecon
structionventilationfrequency.Furthermore ,theIHPO - XDFmodelwasanalyzedusingtheSHapleyAdditiveex
Planations (SHAP)forinterpretability,whichcouldminethekeyfeaturesthataffectedthepredictionresultsofthe
constructionventilationfrequencyandenhancethecredibilityoftheventilationfrequencypredictionmodel.The
casestudyshowedthatcomparedwiththeXGBoost - improvedDFmodel,thetraditionalDF,theGradientBoosting
DecisionTree (GBDT),andtheDecisionTree(DT)models,theproposedmodelachievedarespectiveimprove
mentof3.48%,5.01%,13.13%,and13.48% inpredictionaccuracy,andtheIHPO - XDFmodelshowedthe
smallestdecreaseinpredictionaccuracyintheabnormalvalueenvironment,atonly2.17%,demonstratinggood
robustness.Meanwhile ,theSHAP - basedinterpretabilityanalysisofthepredictionresultsindicatedthattheIHPO -
XDFalgorithmhadhighcredibility.
Keywords:undergroundpowerhousecaverngroup;robustpredictionofventilationfrequency;deepforestmodel;
XGBoost;improvedhunter - preyalgorithm;interpretability
(责任编辑:韩 昆)
0
— 1 8 3 —

