由于滚动轴承实际工作环境恶劣,含标签故障样本数据严重缺乏,不足以建立准确的预测模型。支持矩阵机(support matrix machine,SMM)作为一种新的模式识别方法,可以获得良好的分类效果,但其仍对小样本分析具有局限性。基于此,提出一种迁...由于滚动轴承实际工作环境恶劣,含标签故障样本数据严重缺乏,不足以建立准确的预测模型。支持矩阵机(support matrix machine,SMM)作为一种新的模式识别方法,可以获得良好的分类效果,但其仍对小样本分析具有局限性。基于此,提出一种迁移最小二乘支持矩阵机(transfer least square support matrix machine,TLSSMM)分类方法。在TLSSMM分类过程中,利用源域样本训练得到近似目标域的预测模型,并通过目标域少量含标签样本微调源域的训练模型以更新得到新模型。同时,采用最小二乘损失来约束目标函数,使其由不等式转换为等式,只需求解一组线性方程即可获得结果,大大提升分类效率。选择两种不同的滚动轴承故障数据对所提方法进行验证,实验结果表明,TLSSMM方法具有优异的分类性能。展开更多
AUC(area under the ROC curve)优化问题的损失函数由来自不同类别的样本对构成,这使得依赖于损失函数之和的目标函数与训练样本数二次相关,不能直接使用传统在线学习方法求解.当前的在线AUC优化算法聚焦于在求解过程中避免直接计算所...AUC(area under the ROC curve)优化问题的损失函数由来自不同类别的样本对构成,这使得依赖于损失函数之和的目标函数与训练样本数二次相关,不能直接使用传统在线学习方法求解.当前的在线AUC优化算法聚焦于在求解过程中避免直接计算所有的损失函数,以减小问题的规模,实现在线AUC优化.针对以上问题提出了一种AUC优化的新目标函数,该目标函数仅与训练样本数线性相关;理论分析表明:最小化该目标函数等价于最小化由L2正则化项和最小二乘损失函数组成的AUC优化的目标函数.基于新的目标函数,提出了在线AUC优化的线性方法(linear online AUC maximization,LOAM);根据不同的分类器更新策略,给出2种算法LOAMILSC和LOAMAda.实验表明:与原有方法相比,LOAMILSC算法获得了更优的AUC性能,而对于实时或高维学习任务,LOAMAda算法更加高效.展开更多
To predict the economic loss of crops caused by acid rain,we used partial least squares(PLS) regression to build a model of single dependent variable -the economic loss calculated with the decrease in yield related to...To predict the economic loss of crops caused by acid rain,we used partial least squares(PLS) regression to build a model of single dependent variable -the economic loss calculated with the decrease in yield related to the pH value and levels of Ca2+,NH4+,Na+,K+,Mg2+,SO42-,NO3-,and Cl-in acid rain. We selected vegetables which were sensitive to acid rain as the sample crops,and collected 12 groups of data,of which 8 groups were used for modeling and 4 groups for testing. Using the cross validation method to evaluate the performace of this prediction model indicates that the optimum number of principal components was 3,determined by the minimum of prediction residual error sum of squares,and the prediction error of the regression equation ranges from -2.25% to 4.32%. The model predicted that the economic loss of vegetables from acid rain is negatively corrrelated to pH and the concentrations of NH4+,SO42-,NO3-,and Cl-in the rain,and positively correlated to the concentrations of Ca2+,Na+,K+ and Mg2+. The precision of the model may be improved if the non-linearity of original data is addressed.展开更多
文摘由于滚动轴承实际工作环境恶劣,含标签故障样本数据严重缺乏,不足以建立准确的预测模型。支持矩阵机(support matrix machine,SMM)作为一种新的模式识别方法,可以获得良好的分类效果,但其仍对小样本分析具有局限性。基于此,提出一种迁移最小二乘支持矩阵机(transfer least square support matrix machine,TLSSMM)分类方法。在TLSSMM分类过程中,利用源域样本训练得到近似目标域的预测模型,并通过目标域少量含标签样本微调源域的训练模型以更新得到新模型。同时,采用最小二乘损失来约束目标函数,使其由不等式转换为等式,只需求解一组线性方程即可获得结果,大大提升分类效率。选择两种不同的滚动轴承故障数据对所提方法进行验证,实验结果表明,TLSSMM方法具有优异的分类性能。
文摘AUC(area under the ROC curve)优化问题的损失函数由来自不同类别的样本对构成,这使得依赖于损失函数之和的目标函数与训练样本数二次相关,不能直接使用传统在线学习方法求解.当前的在线AUC优化算法聚焦于在求解过程中避免直接计算所有的损失函数,以减小问题的规模,实现在线AUC优化.针对以上问题提出了一种AUC优化的新目标函数,该目标函数仅与训练样本数线性相关;理论分析表明:最小化该目标函数等价于最小化由L2正则化项和最小二乘损失函数组成的AUC优化的目标函数.基于新的目标函数,提出了在线AUC优化的线性方法(linear online AUC maximization,LOAM);根据不同的分类器更新策略,给出2种算法LOAMILSC和LOAMAda.实验表明:与原有方法相比,LOAMILSC算法获得了更优的AUC性能,而对于实时或高维学习任务,LOAMAda算法更加高效.
基金Funded by the Natural Basic Research Program of China under the grant No. 2005CB422207.
文摘To predict the economic loss of crops caused by acid rain,we used partial least squares(PLS) regression to build a model of single dependent variable -the economic loss calculated with the decrease in yield related to the pH value and levels of Ca2+,NH4+,Na+,K+,Mg2+,SO42-,NO3-,and Cl-in acid rain. We selected vegetables which were sensitive to acid rain as the sample crops,and collected 12 groups of data,of which 8 groups were used for modeling and 4 groups for testing. Using the cross validation method to evaluate the performace of this prediction model indicates that the optimum number of principal components was 3,determined by the minimum of prediction residual error sum of squares,and the prediction error of the regression equation ranges from -2.25% to 4.32%. The model predicted that the economic loss of vegetables from acid rain is negatively corrrelated to pH and the concentrations of NH4+,SO42-,NO3-,and Cl-in the rain,and positively correlated to the concentrations of Ca2+,Na+,K+ and Mg2+. The precision of the model may be improved if the non-linearity of original data is addressed.
