15··基于SA-PSO算法优化LS-SVM的基坑土层等效参数反演*曹净唐斌懿李豪(昆明理工大学建筑工程学院,云南昆明650500)摘要针对基坑支护结构位移与土层参数具有小样本及非线性的特征,提出一种以模拟退火(SA)算法与粒子群(PSO)算法混合优化最小二乘支持向量机(LSSVM)的位移反分析方法。一是通过均匀试验构造学习与测试样本,运用SA-PSO混合算法对最小二乘支持向量机进行参数寻优,寻找模型最优参数组合,并建立最小二乘支持向量机非线性回归模型;二是构建预测位移与实测位移间的目标函数,运用SA-PSO混合算法迭代寻优基坑土层参数值。应用于昆明某实际基坑工程中,反演结果表明此方法具有一定的可行性。关键词基坑土层均匀试验模拟退火-粒子群算法最小二乘支持向量机参数反演OptimizationofLS-SVMbasedonSA-PSOalgorithmforinversionofequivalentparametersoffoundationsoillayersCAOJingTANGBinyiLIHao(FacultyofCivilEngineeringandMechanics,KunmingUniversityofScienceandTechnology,KunmingYunnan650500,China)AbstractForthecharacteristicsofsmallsamplesandnonlinearityofthedisplacementandsoilparametersofthefoundationsupportstructure,adisplacementinverseanalysismethodofleastsquaressupportvectormachine(LSSVM)isproposedwiththehybridsimulatedannealing(SA)algorithmandparticleswarm(PSO)algorithm.Thefirstistoconstructlearningandtestingsamplesthroughuniformexperiments,applytheSA-PSOhybridalgorithmtotheleastsquaressupportvectormachineforparameteroptimization,findtheoptimalcombinationofmodelparame-ters,andestablishtheleastsquaressupportvectormachinenonlinearregressionmodel;thesecondistoconstructtheobjectivefunctionbetweenpredictedandmeasureddisplacements,andapplytheSA-PSOhybridalgorithmtoitera-tivelyoptimizetheparametervaluesofthefoundationpitsoillayer.AppliedtoapracticalfoundationpitprojectinKunming,theinversionresultsshowthatthismethodhascertainfeasibility.Keywordsfoundationpitsoiluniformitytestsimulatedannealing-particleswarmalgorithmleastsquaressup-portvectormachineparameterinversion0引言岩土参数反算可视为对应目标函数寻求最优解,而岩土本身是复杂的、具有区域性和时空效应的材料,使得在多数情况下的目标函数是复杂非线性的函...