传统支持向量机分类过程的计算量和支持向量的个数成正比,当支持向量较多时,其分类过程的计算比较耗时。该文基于支持向量的稀疏性,证明了对支持向量压缩时,收紧新的快速决策函数和原始决策函数之间的误差等价于在样本空间对原始支持向...传统支持向量机分类过程的计算量和支持向量的个数成正比,当支持向量较多时,其分类过程的计算比较耗时。该文基于支持向量的稀疏性,证明了对支持向量压缩时,收紧新的快速决策函数和原始决策函数之间的误差等价于在样本空间对原始支持向量进行K均值聚类操作,据此提出了一种约简支持向量的快速分类算法FD-SVM(Fast Decision algorithm of Support Vector Machine),该算法首先对原始的支持向量进行特定比例的K均值聚类操作,聚类的中心为约简后新的支持向量,按照分类误差最小的原则构建优化模型,用二次规划方法求解得到新的支持向量的系数。标准数据集上的实验表明,保持分类精度的损失在统计意义上不明显的前提下,FD-SVM可以有效压缩支持向量的数量,提高分类速度。展开更多
对电池特性的深刻认识是电池应用研究的重要基础,而弛豫时间分布(distribution of relaxation times,DRT)法是解析电池阻抗谱(electrochemical impedance spectroscopy,EIS)、提取电极过程动力学信息和电池建模的有效手段。然而,DRT函...对电池特性的深刻认识是电池应用研究的重要基础,而弛豫时间分布(distribution of relaxation times,DRT)法是解析电池阻抗谱(electrochemical impedance spectroscopy,EIS)、提取电极过程动力学信息和电池建模的有效手段。然而,DRT函数的求解是一个典型的不适定问题,经典的数值积分方法无法保证解的存在性或唯一性。首先采用分段线性插值近似连续的DRT函数;再通过正则化方法改善问题的不适定性,将DRT函数的求解归结为严格的凸二次规划(quadratic programming,QP)问题;进而运用有效集法(active set method,ASM)得到DRT函数的最优近似解。基于该方法解析液态金属电池的阻抗谱,并简要分析其内阻特性。研究结果表明:该方法为全局收敛,收敛速度快,计算精度高;得到的DRT函数近似解既精确、稳定,又具有明确的物理意义。在电池机理分析和建模中,该方法具有显著的潜在应用价值。展开更多
A compliant landing strategy for a trotting quadruped robot on unknown rough terrains based on contact force control is presented. Firstly, in order to lower the disturbance caused by the landing impact force, a landi...A compliant landing strategy for a trotting quadruped robot on unknown rough terrains based on contact force control is presented. Firstly, in order to lower the disturbance caused by the landing impact force, a landing phase is added between the swing phase and the stance phase, where the desired contact force is set as a small positive constant. Secondly, the joint torque optimization of the stance legs is formulated as a quadratic programming(QP) problem subject to equality and inequality/bound constraints. And a primal-dual dynamical system solver based on linear variational inequalities(LVI) is applied to solve this QP problem. Furthermore, based on the optimization results, a hybrid motion/force robust controller is designed to realize the tracking of the contact force, while the constraints of the stance feet landing angles are fulfilled simultaneously. Finally, the experiments are performed to validate the proposed methods.展开更多
文摘传统支持向量机分类过程的计算量和支持向量的个数成正比,当支持向量较多时,其分类过程的计算比较耗时。该文基于支持向量的稀疏性,证明了对支持向量压缩时,收紧新的快速决策函数和原始决策函数之间的误差等价于在样本空间对原始支持向量进行K均值聚类操作,据此提出了一种约简支持向量的快速分类算法FD-SVM(Fast Decision algorithm of Support Vector Machine),该算法首先对原始的支持向量进行特定比例的K均值聚类操作,聚类的中心为约简后新的支持向量,按照分类误差最小的原则构建优化模型,用二次规划方法求解得到新的支持向量的系数。标准数据集上的实验表明,保持分类精度的损失在统计意义上不明显的前提下,FD-SVM可以有效压缩支持向量的数量,提高分类速度。
文摘对电池特性的深刻认识是电池应用研究的重要基础,而弛豫时间分布(distribution of relaxation times,DRT)法是解析电池阻抗谱(electrochemical impedance spectroscopy,EIS)、提取电极过程动力学信息和电池建模的有效手段。然而,DRT函数的求解是一个典型的不适定问题,经典的数值积分方法无法保证解的存在性或唯一性。首先采用分段线性插值近似连续的DRT函数;再通过正则化方法改善问题的不适定性,将DRT函数的求解归结为严格的凸二次规划(quadratic programming,QP)问题;进而运用有效集法(active set method,ASM)得到DRT函数的最优近似解。基于该方法解析液态金属电池的阻抗谱,并简要分析其内阻特性。研究结果表明:该方法为全局收敛,收敛速度快,计算精度高;得到的DRT函数近似解既精确、稳定,又具有明确的物理意义。在电池机理分析和建模中,该方法具有显著的潜在应用价值。
基金Project(61473304)supported by the National Natural Science Foundation of ChinaProject(2015AA042202)supported by Hi-tech Research and Development Program of China
文摘A compliant landing strategy for a trotting quadruped robot on unknown rough terrains based on contact force control is presented. Firstly, in order to lower the disturbance caused by the landing impact force, a landing phase is added between the swing phase and the stance phase, where the desired contact force is set as a small positive constant. Secondly, the joint torque optimization of the stance legs is formulated as a quadratic programming(QP) problem subject to equality and inequality/bound constraints. And a primal-dual dynamical system solver based on linear variational inequalities(LVI) is applied to solve this QP problem. Furthermore, based on the optimization results, a hybrid motion/force robust controller is designed to realize the tracking of the contact force, while the constraints of the stance feet landing angles are fulfilled simultaneously. Finally, the experiments are performed to validate the proposed methods.
