为提高月径流时间序列预测精度,改进最小二乘孪生支持向量回归机(Least Squares Twin Support Vector Regression,LSTSVR)性能,首次基于高斯核函数、多项式核函数、线性核函数构建混合核函数,提出4种季节趋势分解(Seasonal and Trend de...为提高月径流时间序列预测精度,改进最小二乘孪生支持向量回归机(Least Squares Twin Support Vector Regression,LSTSVR)性能,首次基于高斯核函数、多项式核函数、线性核函数构建混合核函数,提出4种季节趋势分解(Seasonal and Trend decomposition using Loess,STL)-小波包变换(Wavelet Packet Transform,WPT)-裂狐优化(Rüppell's Fox Optimizer,RFO)算法-混合核最小二乘孪生支持向量回归机(Hybrid Kernel Least Squares Twin Support Vector Regression,HLSTSVR)模型,并构建STL-WPT-RFO-LSTSVR、STL-WPT-RFO-混合核最小二乘支持向量回归机(Hybrid Kerllel Least Squares Twin Suppart Vector Regression,HLSSVR)、STL-WPT-RFO-最小二乘支持向量回归机(Least Squares Support Vector Regression,LSSVR)等17种对比分析模型,通过云南省高桥、凤屯水文站月径流时间序列预测实例对21种模型进行验证。首先利用STL-WPT二次分解技术对月径流序列进行分解处理,合理划分训练集和验证集;然后基于高斯核函数、多项式核函数、线性核函数,采用“三三”线性组合和“两两”线性组合的方式构建4种混合核函数对月径流分解分量进行空间映射;最后利用RFO寻优HLSTSVR/LSTSVR/HLSSVR/LSSVR最佳超参数,利用最佳超参数建立21种模型对实例月径流序列各分解分量进行训练、预测和重构。结果表明:①4种STL-WPT-RFO-HLSTSVR模型能适应不同尺度的月径流数据分布,具有较好的模型性能和较小的预测误差,其中STL-WPT-RFO-HLSTSVR(高斯+多项式+线性)模型对高桥、凤屯站月径流预测的平均绝对百分比误差MAPE分别为2.85%、2.19%,决定系数R2均为0.9994,预测精度最高、效果最好;②混合核函数兼顾了不同核函数优势,能在模型复杂度与泛化能力之间取得平衡,显著提升模型性能和预测精度;③STL-WPT二次分解技术能有效解决复杂时间序列的非平稳性、非线性和多尺度特征,较STL更具分解优势;④组合模型融合了STL-WPT、RFO和HLSTSVR优点,具有较好的普适性和参考价值。展开更多
Twin support vector machine(TWSVM)is a new development of support vector machine(SVM)algorithm.It has the smaller computation scale and the stronger ability to cope with unbalanced problems.In this paper,TWSVM is intr...Twin support vector machine(TWSVM)is a new development of support vector machine(SVM)algorithm.It has the smaller computation scale and the stronger ability to cope with unbalanced problems.In this paper,TWSVM is introduced into aircraft engine gas path fault diagnosis.The generalization capacity of Gauss kernel function usually used in TWSVM is relatively weak.So a mixed kernel function is used to improve performance to ensure that the TWSVM algorithm can better balance a strong generalization ability and a good learning ability.Experimental results prove that the cross validation training accuracy of TWSVM using the mixed kernel function averagely increases 2%.Grid search is usually applied in parameter optimization of TWSVM,but it heavily depends on experience.Therefore,the hybrid particle swarm algorithm is introduced.It can intelligently and rapidly find the global optimum.Experiments prove that its training accuracy is better than that of the classical particle swarm algorithm by 5%.展开更多
针对最小二乘孪生支持向量机受误差值影响大,对噪声样本敏感及核函数、核参数选择困难等问题,提出一种Critic特征加权的多核最小二乘孪生支持向量机(Multi-Kernel Least-Squares Twin Support Vector Machine based on Critic weighted,...针对最小二乘孪生支持向量机受误差值影响大,对噪声样本敏感及核函数、核参数选择困难等问题,提出一种Critic特征加权的多核最小二乘孪生支持向量机(Multi-Kernel Least-Squares Twin Support Vector Machine based on Critic weighted,CMKLSTSVM)分类方法。首先,CMKLSTSVM使用Critic法赋予特征权重,反映不同特征间重要性差异,降低冗余特征及噪声样本影响。其次,根据混合多核学习策略构造了一种新的多核权重系数确定方法。该方法通过基核与理想核间的混合核对齐值判断核函数相似程度,确定权重系数,可以合理地组合多个核函数,最大程度地发挥不同核函数的映射能力。最后,采用加权求和的方式将特征权重与核权重进行统一并构造多核结构,使数据表达更全面,提高模型灵活性。在UCI数据集上的对比实验表明,CMKLSTSVM的分类准确率优于单核结构的SVM(support vector machine)算法,同时在高光谱图像上的对比实验反映了CMKLSTSVM对于包含噪声的真实分类问题的有效性。展开更多
文摘为提高月径流时间序列预测精度,改进最小二乘孪生支持向量回归机(Least Squares Twin Support Vector Regression,LSTSVR)性能,首次基于高斯核函数、多项式核函数、线性核函数构建混合核函数,提出4种季节趋势分解(Seasonal and Trend decomposition using Loess,STL)-小波包变换(Wavelet Packet Transform,WPT)-裂狐优化(Rüppell's Fox Optimizer,RFO)算法-混合核最小二乘孪生支持向量回归机(Hybrid Kernel Least Squares Twin Support Vector Regression,HLSTSVR)模型,并构建STL-WPT-RFO-LSTSVR、STL-WPT-RFO-混合核最小二乘支持向量回归机(Hybrid Kerllel Least Squares Twin Suppart Vector Regression,HLSSVR)、STL-WPT-RFO-最小二乘支持向量回归机(Least Squares Support Vector Regression,LSSVR)等17种对比分析模型,通过云南省高桥、凤屯水文站月径流时间序列预测实例对21种模型进行验证。首先利用STL-WPT二次分解技术对月径流序列进行分解处理,合理划分训练集和验证集;然后基于高斯核函数、多项式核函数、线性核函数,采用“三三”线性组合和“两两”线性组合的方式构建4种混合核函数对月径流分解分量进行空间映射;最后利用RFO寻优HLSTSVR/LSTSVR/HLSSVR/LSSVR最佳超参数,利用最佳超参数建立21种模型对实例月径流序列各分解分量进行训练、预测和重构。结果表明:①4种STL-WPT-RFO-HLSTSVR模型能适应不同尺度的月径流数据分布,具有较好的模型性能和较小的预测误差,其中STL-WPT-RFO-HLSTSVR(高斯+多项式+线性)模型对高桥、凤屯站月径流预测的平均绝对百分比误差MAPE分别为2.85%、2.19%,决定系数R2均为0.9994,预测精度最高、效果最好;②混合核函数兼顾了不同核函数优势,能在模型复杂度与泛化能力之间取得平衡,显著提升模型性能和预测精度;③STL-WPT二次分解技术能有效解决复杂时间序列的非平稳性、非线性和多尺度特征,较STL更具分解优势;④组合模型融合了STL-WPT、RFO和HLSTSVR优点,具有较好的普适性和参考价值。
基金supported by the Fundamental Research Funds for the Central Universities(No.NS2016027)
文摘Twin support vector machine(TWSVM)is a new development of support vector machine(SVM)algorithm.It has the smaller computation scale and the stronger ability to cope with unbalanced problems.In this paper,TWSVM is introduced into aircraft engine gas path fault diagnosis.The generalization capacity of Gauss kernel function usually used in TWSVM is relatively weak.So a mixed kernel function is used to improve performance to ensure that the TWSVM algorithm can better balance a strong generalization ability and a good learning ability.Experimental results prove that the cross validation training accuracy of TWSVM using the mixed kernel function averagely increases 2%.Grid search is usually applied in parameter optimization of TWSVM,but it heavily depends on experience.Therefore,the hybrid particle swarm algorithm is introduced.It can intelligently and rapidly find the global optimum.Experiments prove that its training accuracy is better than that of the classical particle swarm algorithm by 5%.
文摘针对最小二乘孪生支持向量机受误差值影响大,对噪声样本敏感及核函数、核参数选择困难等问题,提出一种Critic特征加权的多核最小二乘孪生支持向量机(Multi-Kernel Least-Squares Twin Support Vector Machine based on Critic weighted,CMKLSTSVM)分类方法。首先,CMKLSTSVM使用Critic法赋予特征权重,反映不同特征间重要性差异,降低冗余特征及噪声样本影响。其次,根据混合多核学习策略构造了一种新的多核权重系数确定方法。该方法通过基核与理想核间的混合核对齐值判断核函数相似程度,确定权重系数,可以合理地组合多个核函数,最大程度地发挥不同核函数的映射能力。最后,采用加权求和的方式将特征权重与核权重进行统一并构造多核结构,使数据表达更全面,提高模型灵活性。在UCI数据集上的对比实验表明,CMKLSTSVM的分类准确率优于单核结构的SVM(support vector machine)算法,同时在高光谱图像上的对比实验反映了CMKLSTSVM对于包含噪声的真实分类问题的有效性。
