第14卷第4期2023年2月黑龙江科学HEILONGJIANGSCIENCEVol.14Feb.2023基于BP-MCS结构可靠度模型的样本分析章浩龙,刘洪君(广东理工学院建设学院,广东肇庆526100)摘要:针对结构可靠度计算,以BP神经网络代替一般响应面的多项式函数、结合MCS蒙特卡洛算法组合的模型较为常见。该方法充分发挥了BP神经网络与MCS的优点,利用神经网络泛化能力解决了隐式功能函数的结构失效概率的求解难题。但目前大多研究忽视了BP-MCS组合模型的样本容量选取问题,对一个已有显式功能函数的岩质边坡数值算例进行研究,发现BP神经网络训练样本与MCS算法抽样的选取对BP-MCS模型计算精度有重要的影响,分析结果可为实际工程结构可靠度计算提供建议。关键词:结构可靠度;失效概率;样本选取;神经网络;蒙特卡洛算法中图分类号:TP183文献标志码:A文章编号:1674-8646(2023)04-0022-04SampleAnalysisBasedonBP-MCSModelforStructureReliabilityZhangHaolong,LiuHongjun(ConstructionInstituteofGuangdongTechnologyCollege,Zhaoqing526100,China)Abstract:Numerousstudiesfocusonthecalculationofstructurereliability.ItiscommontousetheBPneuralnetworkinsteadofthepolynomialfunctionofthegeneralresponsesurfacecombinedwiththeMonteCarloalgorithm,toformacombinedmodel.AdvantagesofBPneuralnetworkandMCSarehighlybroughtintothismodelforusingthegeneralizationabilityofneuralnetworksolvingtheproblemofansweringthestructurefailureprobabilityunderimplicitfunction.However,thesamplesizeselectionoftheBP-MCScombinedmodelhasbeenignoredinmostresearchescurrently.Inthispaper,itisfoundthatthesampleselectionoftheBPneuralnetworktrainingandtheMCSalgorithmhasasignificantimpactincalculationaccuracyoftheBP-MCSmodelthrougharockslopenumericalexamplewithanexplicitfunction.Theanalysiscanprovidesomedirectionalguidanceforthereliabilitycalculationofactualengineeringstructures.Keywords:Structurereliability;Failureprobability;Sampleselection;Neuralnetwork;Montecarloalgorithm收稿日期:2022-12-02作者简介:刘洪君(1995-),男,硕士研究生。通讯作者:章浩龙(1997-),男,硕士研究生。E-mail:1429482212@qq.com。0引言结构可靠度与其安全性密切相关,是某结构在规定时间和规定条件下完成预定功能的概率。对应而言,结构失效概...