基于自适应权重的函数型数据聚类方法研究

基于有限维离散数据的传统聚类分析并不能直接用于函数型数据的分类挖掘。本文针对函数型数据的稀疏性和无穷维特殊性展开讨论,在综合剖析现有函数型聚类方法优势与不足的基础上,依据聚类指标的信息量差异重构加权主成分距离为函数相似性测度,提出了一种函数型数据的自适应权重聚类分析。相对同类函数型聚类算法,新方法的核心优势在于:(1)自适应赋权的距离函数体现了聚类指标分类效率的差异,并且有充分的理论基础保证其必要性和客观合理性;(2)基于有限维离散数据的聚类实现了无限维连续函数的聚类,能够显著降低计算成本。实证检验表明,新方法的分类正确率明显提高,能够有效解决传统聚类算法极端情形下的失效问题,有着复杂函数型数据分类问题下的灵活性和普遍适用性。     

基于变分贝叶斯优化的近邻采样 PF-SLAM 算法

针对移动机器人同时定位与地图构建（SLAM）中观测噪声随时间变化及粒子滤波(PF)中粒子多样性易丧失问题,提出基于变分贝叶斯优化的近邻采样PF-SLAM算法。采用高斯混合模型对时变的观测噪声建模,使用变分贝叶斯方法,迭代估算出混合模型中的未知参数;同时根据粒子权重将粒子划分为保留粒子和调整粒子,通过两种粒子间的近邻位置分布关系优化调整粒子位置,在处理时变观测噪声同时,解决粒子多样性丧失问题,使得优化的粒子集更好地表示机器人位置概率分布。实验表明,改进算法与传统PF-SLAM算法相比,定位与建图误差降低76%,较期望最大化算法下的定位与建图误差降低了54%,进一步验证了所提算法的可行性与有效性,为移动机器人同时定位与建图提供一定参考。     

Optimization designed large-stroke MEMS micromirror for adaptive optics

A novel micromirror based on the PolyMUMPs process is designed and presented. The hexagonal micromirror with a diameter of 450 μm consists of three supporting bilayer cantilevers and a mirror plate. The bilayer cantilevers, formed with a polysilicon layer and a gold layer, elevate the mirror plate according to residual stress-induced bending. Both analytical and finite element analysis (FEA) models are built to calculate the elevated height of the free end of the cantilever. The analytical solution is in accordance with the FEA simulation results, with longitudinal stresses applied only. Results of a three-dimensional (3D) simulation with two direction stresses applied also show the elevated height to be proportional to the width of the cantilever and the length of the gold layer. Due to the torque of the joint, the elevated heights of the two kinds of cantilevers assembled with the mirror plates are much smaller than those of the free end of the cantilevers. Both micromirrors with different cantilevers are fabricated. The elevated heights of the fabricated micromirrors are measured using Veeco optical profiler, which show good coincidence with simulation results.     

THEORY OF DIELECTRIC ELASTOMERS

In response to a stimulus, a soft material deforms, and the deformation provides a function. We call such a material a soft active material （SAM）. This review focuses on one class of soft active materials： dielectric elastomers. When a membrane of a dielectric elastomer is subject to a voltage through its thickness, the membrane reduces thickness and expands area, possibly straining over 100%. The dielectric elastomers are being developed as transducers for broad applications, including soft robots, adaptive optics, Braille displays, and electric generators. This paper reviews the theory of dielectric elastomers, developed within continuum mechanics and thermodynamics, and motivated by molecular pictures and empirical observations. The theory couples large deformation and electric potential, and describes nonlinear and nonequilibrium behavior, such as electromechanical instability and viscoelasticity. The theory enables the finite element method to simulate transducers of realistic configurations, predicts the efficiency of electromechanical energy conversion, and suggests alternative routes to achieve giant voltage-induced deformation. It is hoped that the theory will aid in the creation of materials and devices.     

Control of a Unified Chaotic System via Single Variable Feedback

Based on the LaSalle invariance principle, we propose a simple adaptive-feedback for controlling the unified chaotic system. We show explicitly with numerical proofs that our method can easily achieve the control of chaos in the unified chaotic system using only a single variable feedback. The present controller, to our knowledge, is the simplest control scheme for controlling a unified chaotic system.     

改进遗传算法在天然气液化工艺流程优化中的应用

天然气液化工艺流程的设计优化中所涉及的设备、运行参数以及物流的物性参数众多,加之各热力学方程高度非线性这些特点,国内外一些学者将遗传算法这一优秀算法引入到天然气液化流程的优化设计中。但标准遗传算法在天然气液化工艺的优化设计中容易陷入局部最优的陷阱,且收敛的速度及精度还有待提升。本文根据天然气液化工艺优化设计中待优化参数多,耗时长的特点对标准遗传算法进行改进,改进后的遗传算法在提升全局搜索能力的同时还增强了局部搜索能力,对天然气液化流程的优化设计具有很好的适用性。     

闪光照相CCD图像的自适应中值滤波方法

中值滤波是一种在去除噪声的同时能较好地保护图像边缘细节的非线性图像处理方法。为了滤除闪光照相CCD图像中的脉冲噪声，同时能更好地保护图像边缘，提出了一种改进的自适应中值滤波方法。该方法采用局部中值和局部方差作为判断噪声点的阈值，实现了局部自适应的中值滤波，克服了传统自适应中值滤波方法的缺点，对椒盐噪声和随机脉冲噪声均有较好的滤波效果。实验结果表明，该方法消除图像脉冲噪声十分有效，对闪光照相CCD图像的处理结果也较好。     

遗传算法用于多组分拉曼光谱定量分析的研究

拉曼光谱是一种分子散射光谱,利用物质的特征拉曼光谱,可实现对其结构、成分、浓度等的检测.多组分物质拉曼光谱的定量分析是一个有待深入研究的课题.采用遗传算法程序对10组分的混合矿石样品的拟合拉曼光谱数据进行了分析,对同一条件下算法多次计算的结果和不同条件下算法的表现进行了讨论,分析了算法设计和基本参数的不同对算法收敛速度以及对计算结果的精确性和稳定性的影响,验证了遗传算法用于多组分样品拉曼光谱数据分析的可行性、有效性和精确性.     

父代种群参与竞争遗传算法几乎必然收敛

熟知,标准遗传算法如不采用“杰出者记录策略”则必不收敛。本文发现：允许父代种群参与竞争是标准遗传算法几乎必然收敛的条件。特别地,我们运用鞅收敛定理证明：允许父代种群参与竞争型遗传算法能以概率1确保在限步内达到全局最优解,且收敛与种群规模无关。所获结果对该类遗传算法的应用奠定了可靠基础。     

掺镱双包层光纤放大器分布泵浦方式下的优化算法

 利用平均反转率迭代算法计算掺镱双包层光纤放大器分布泵浦方式下的稳态速率方程组，并采用遗传算法对分布泵浦功率的大小和每段光纤长度同时进行优化。评估函数中引入了每段光纤中最高温度的标准方差，以确保每段光纤中的最高工作温度是相同的。通过优化，7段泵浦的最高温度和标准方差分别为126.34 ℃和1.95 ℃ ，8段泵浦条件下的最高温度和标准方差分别为119.76 ℃和2.12 ℃。而未经优化的7段泵浦的最高温度和标准方差分别为147.12 ℃和21.37 ℃，8段泵浦条件下的最高温度和标准方差分别为139.95 ℃和20.83 ℃。计算结果表明：泵浦方式的优化降低了最高温度和标准方差，实现了光纤中温度分布的均匀性，并且通过增加泵浦段数可以进一步降低最高温度和平坦温度分布。