高精度且鲁棒的预测模型建立高度依赖于样本数据的大小、多样性和分布;日益积累的文献数据为获得大量的多样性样本数据提供了可能。以SLM-ed IN 718合金的相对密度(RD)为研究对象,针对从文献中挖掘的激光功率P、扫描速度V、扫描间距HS...高精度且鲁棒的预测模型建立高度依赖于样本数据的大小、多样性和分布;日益积累的文献数据为获得大量的多样性样本数据提供了可能。以SLM-ed IN 718合金的相对密度(RD)为研究对象,针对从文献中挖掘的激光功率P、扫描速度V、扫描间距HS和铺粉厚度LT与RD样本数据存在缺失参数和分布不均问题,采用最大期望化(EM)算法对缺失参数进行补齐;采用带有梯度惩罚的WGAN算法(WGAN-GP)对数据稀疏的低RD区间生成虚拟样本数据。然后,分别基于补齐文献数据和补充虚拟数据,采用常青藤算法优化的随机森林(IVYA-RF)构建了RD预测模型,并对模型预测精度进行了定量评估和实验验证。结果表明:基于补充虚拟数据集构建的IVYA-RF模型II比基于补齐文献数据集构建的IVYA-RF模型I具有更好的预测精度,其原因主要来源于在低RD区间生成虚拟数据后,使建模样本数据的分布均匀性得到改善,这也是数据增强与机器学习相结合的意义所在。对新实验数据的验证取得了满意的预测精度,其中,IVYA-RF模型I验证结果的统计学参数R2(决定系数)、RMSE(均方根误差)、MAE(平均绝对误差)和MRE(平均相对误差)分别达到了0.891、1.352%、0.915%和0.98%;IVYA-RF模型II验证结果的R2增大至0.956,RMSE、MAE和MRE分别减小至0.833%、0.687%和0.71%,同样显示出后者比前者具有更好的预测精度。实验验证结果表明,所构建的RD预测模型具有较好的鲁棒性,从而具备了较好的工程应用价值。展开更多
Seismic force reduction factor(SFRF) spectra of shear-type multi-degree-of-freedom(MDOF) structures are investigated. The modified Clough model, capable of considering the strength-degradation/hardening and stiffnessd...Seismic force reduction factor(SFRF) spectra of shear-type multi-degree-of-freedom(MDOF) structures are investigated. The modified Clough model, capable of considering the strength-degradation/hardening and stiffnessdegradation, is adopted. The SFRF mean spectra using 102 earthquake records on a typical site soil type(type C) are constructed with the period abscissa being divided into three period ranges to maintain the peak features at the two sitespecific characteristic periods. Based on a large number of results, it is found that the peak value of SFRF spectra may also exist for MDOF, induced by large high-mode contributions to elastic base shear, besides the mentioned two peak values. The variations of the stiffness ratio λk and the strength ratio λF of the top to bottom story are both considered. It is found that the SFRFs for λF ≤λk are smaller than those for λF > λk. A SFRF modification factor for MDOF systems is proposed with respect to SDOF. It is found that this factor is significantly affected by the story number and ductility. With a specific λF(= λk0.75), SFRF mean spectra are constructed and simple solutions are presented for MDOF systems. For frames satisfying the strong column/weak beam requirement, an approximate treatment in the MDOF shear-beam model is to assign a post-limit stiffness 15%-35% of the initial stiffness to the hysteretic curve. SFRF spectra for MDOF systems with 0.2 and 0.3 times the post-limit stiffness are remarkably larger than those without post-limit stiffness. Thus, the findings that frames with beam hinges have smaller ductility demand are explained through the large post-limit stiffness.展开更多
文摘高精度且鲁棒的预测模型建立高度依赖于样本数据的大小、多样性和分布;日益积累的文献数据为获得大量的多样性样本数据提供了可能。以SLM-ed IN 718合金的相对密度(RD)为研究对象,针对从文献中挖掘的激光功率P、扫描速度V、扫描间距HS和铺粉厚度LT与RD样本数据存在缺失参数和分布不均问题,采用最大期望化(EM)算法对缺失参数进行补齐;采用带有梯度惩罚的WGAN算法(WGAN-GP)对数据稀疏的低RD区间生成虚拟样本数据。然后,分别基于补齐文献数据和补充虚拟数据,采用常青藤算法优化的随机森林(IVYA-RF)构建了RD预测模型,并对模型预测精度进行了定量评估和实验验证。结果表明:基于补充虚拟数据集构建的IVYA-RF模型II比基于补齐文献数据集构建的IVYA-RF模型I具有更好的预测精度,其原因主要来源于在低RD区间生成虚拟数据后,使建模样本数据的分布均匀性得到改善,这也是数据增强与机器学习相结合的意义所在。对新实验数据的验证取得了满意的预测精度,其中,IVYA-RF模型I验证结果的统计学参数R2(决定系数)、RMSE(均方根误差)、MAE(平均绝对误差)和MRE(平均相对误差)分别达到了0.891、1.352%、0.915%和0.98%;IVYA-RF模型II验证结果的R2增大至0.956,RMSE、MAE和MRE分别减小至0.833%、0.687%和0.71%,同样显示出后者比前者具有更好的预测精度。实验验证结果表明,所构建的RD预测模型具有较好的鲁棒性,从而具备了较好的工程应用价值。
基金Supported by:National Key Research and Development Program of China under Grant No.2016YFC0701201
文摘Seismic force reduction factor(SFRF) spectra of shear-type multi-degree-of-freedom(MDOF) structures are investigated. The modified Clough model, capable of considering the strength-degradation/hardening and stiffnessdegradation, is adopted. The SFRF mean spectra using 102 earthquake records on a typical site soil type(type C) are constructed with the period abscissa being divided into three period ranges to maintain the peak features at the two sitespecific characteristic periods. Based on a large number of results, it is found that the peak value of SFRF spectra may also exist for MDOF, induced by large high-mode contributions to elastic base shear, besides the mentioned two peak values. The variations of the stiffness ratio λk and the strength ratio λF of the top to bottom story are both considered. It is found that the SFRFs for λF ≤λk are smaller than those for λF > λk. A SFRF modification factor for MDOF systems is proposed with respect to SDOF. It is found that this factor is significantly affected by the story number and ductility. With a specific λF(= λk0.75), SFRF mean spectra are constructed and simple solutions are presented for MDOF systems. For frames satisfying the strong column/weak beam requirement, an approximate treatment in the MDOF shear-beam model is to assign a post-limit stiffness 15%-35% of the initial stiffness to the hysteretic curve. SFRF spectra for MDOF systems with 0.2 and 0.3 times the post-limit stiffness are remarkably larger than those without post-limit stiffness. Thus, the findings that frames with beam hinges have smaller ductility demand are explained through the large post-limit stiffness.
