简介:单次脑电分类实验中,采用基于logistic回归的正则化方法来提高分类准确率.首先,提出一种新算法——局部保持投影稀疏logistic回归,将局部保持投影正则项加入到稀疏logistic回归中.该算法旨在保留原始特征空间邻域信息的同时保证结果的稀疏性.然后,利用边界优化法和逐分量迭代算法在训练集上求解权重向量,克服了牛顿一拉夫森法和迭代重加权最小二乘法的局限性.最后,在自步调手指运动数据集上采用十重交叉验证法得到80%的分类准确率,并与稀疏logistic回归的实验结果进行对比,说明局部保持投影正则项有效地保留了对脑电分类有用的信息.
简介:Applyingstiffnessmigrationmethod,a3Dfiniteelementmechanicalmodelisestablishedtosimulatetheexcavationandadvanceprocesses.Byusing3Dnonlinearfiniteelementmethod,thetunnelboringmachine(TBM)excavationprocessisdynamicallysimulatedtoanalyzethestressandstrainfieldstatusofsurroundingrockandsegment.Themaximumtensilestressofsegmentringcausedbytunnelconstructionmainlyliesinarchbottomandpresentszonaldistribution.Thestressincreasesslightlyandlimitedlyinthecourseofexcavation.Themaximumandminimumdisplacementsofsegment,manifestingaszonaldistribution,distributeinarchbottomandvaultrespectively.ThedisplacementsslightlyincreasewiththeadvanceofTBMandgraduallytendtostability.