Smooth constraint inversion technique in genetic algorithms and its application to surface wave study in the Tibetan Plateau

Smooth constraint is important in linear inversion, but it is difficult to apply directly to model parameters in genetic algorithms. If the model parameters are smoothed in iteration, the diversity of models will be greatly suppressed and all the models in population will tend to equal in a few iterations, so the optimal solution meeting requirement can not be obtained. In this paper, an indirect smooth constraint technique is introduced to genetic inversion. In this method, the new models produced in iteration are smoothed, then used as theoretical models in calculation of misfit function, but in process of iteration only the original models are used in order to keep the diversity of models. The technique is effective in inversion of surface wave and receiver function. Using this technique, we invert the phase velocity of Raleigh wave in the Tibetan Plateau, revealing the horizontal variation of S wave velocity structure near the center of the Tibetan Plateau. The results show that the S wave velocity in the north is relatively lower than that in the south. For most paths there is a lower velocity zone with 12-25 km thick at the depth of 15-40 km. The lower velocity zone in upper mantle is located below the depth of 100 km, and the thickness is usually 40-80 km, but for a few paths reach to 100 km thick. Among the area of Ando, Maqi and Ushu stations, there is an obvious lower velocity zone with the lowest velocity of 4.2-4.3 km/s at the depth of 90-230 km. Based on the S wave velocity structures of different paths and former data, we infer that the subduction of the Indian Plate is delimited nearby the Yarlung Zangbo suture zone.

Modal characteristics analysis for a flexible rotor with non-smooth constraint due to intermittent rub-impact

Intermittent rub-impact, during which the contact between rotor and stator is characterized by a ‘‘bouncing＂ or intermittent type of behavior, is one of the most common rubbing forms in rotating machinery. When the intermittent rub-impact occurs, the non-smooth constraint, which is the phenomenon that the system stiffness changes with respect to the state of contact and noncontact, will appear. The paper aims at discovering the possible effects of the non-smooth constraint on the flexible rotor＇s modal characteristics by theoretical and experimental methods. The qualitative description for non-smooth constraint is given for the intermittent rub-impact process, and the dynamic modeling for a rotor system with non-smooth constraint is carried out. Meanwhile, the analysis method is developed by Floquet theory and Hill＇s method to obtain the rotor＇s modal characteristics. The results reveal that the non-smooth constraint produced by the intermittent rubimpact will increase the modal frequencies and critical speeds of the rotor system significantly.Due to the time-varying features of the constraint stiffness, the modal frequencies for the intermittent rub-impact rotor present fluctuant changes with the increase of rotation speed, which is different from the general linear rotor system. The non-smooth constraint is possible to lead the rotor＇s instability, and the rotor＇s instable regions can be expanded significantly for the increase of average constraint stiffness, constraint amplitude and contact time ratio. Non-smooth constraint could also expand the resonance speed and resonance sideband of the rotor system, which sometimes results in amplitude jump phenomenon.

A new smoothing technique for mathematical programs with equilibrium constraints

A kind of mathematical programs with equilibrium constraints （MPEC） is studied. By using the idea of successive approximation, a smoothing nonlinear programming, which is equivalent to the MPEC problem, is proposed. Thereby, it is ensured that some classical optimization methods can be applied for the MPEC problem. In the end, two algorithm models are proposed with the detail analysis of the global convergence.

基于DIoU损失与平滑约束的结构化SVM目标跟踪方法

基于结构化SVM的目标跟踪因其优良的性能而受到了广泛的关注,但是现有方法存在损失函数不精确和模型漂移问题。针对这两个问题,首先提出基于DIoU损失与平滑约束的结构化SVM模型。该模型采用DIoU函数作为损失函数,利用t时刻超平面法向量w_(t)与t-1时刻超平面法向量w_(t-1)差值的L_(2)范数作为平滑约束。其次基于对偶坐标下降原理设计了该模型的求解算法。最后利用提出的基于DIoU损失与平滑约束的结构化SVM实现了一种多尺度目标跟踪方法。对所提出的目标跟踪方法在OTB100和VOT-ST2021数据集上进行了实验验证,实验结果表明:所提出的Scale-DCSSVM在OTB数据集上的跟踪成功率比DeepSRDCF高1.1个百分点,在VOT-ST2021上的EAO比E.T.Track高1.2个百分点。所提方法具有较优的性能。

重力三维密度光滑反演软件设计与实现

采用科学计算语言MATLAB开发一个基于光滑约束的重力三维密度反演软件,包含三维密度反演、重力场正演、数据分析和可视化等功能。测试结果表明,该软件能够反演得到合理的地质体密度分布,具有可靠性和实用性。

基于弹载图像的代价敏感与平滑约束结构化SVM目标跟踪方法

目标自寻的炮弹打击目标时,传统的目标跟踪方法易受混杂背景的影响,产生模型漂移从而导致跟踪失败。近年来,随着结构化SVM技术的发展,基于结构化SVM的目标跟踪方法能够有效解决复杂背景的问题。弹载成像条件下,在结构化SVM目标跟踪方法的基础上,增加代价敏感权重解决弹载图像中背景混杂所引起的正、负样本不平衡问题,同时利用t时刻超平面法向量w_(t)与t-1时刻超平面法向量w_(t-1)差值的L_(2)范数作为平滑约束抑制模型漂移问题。通过基于对偶坐标下降原理设计了模型的求解算法并实现一种多尺度目标跟踪方法Scale-CS_SSVM。在弹载数据集上进行实验验证,结果表明:与Scale-DLSSVM方法相比,Scale-CS_SSVM在跟踪精度和成功率上分别提高了5.5和5.0个百分点,达到了最优的性能。

复杂背景下基于SIFT算法的局部遮挡人脸识别

由于复杂的背景环境中存在大量噪声点、平滑点以及失真点,影响人脸识别效果,提出基于SIFT算法的局部遮挡人脸识别方法。考虑到噪声因素的影响建立尺度不变高斯函数,通过该函数描述原始图像的噪声干扰,查找平滑程度最高的像素点,以像素均值作为参考,对像素点及周围区域实行平滑约束。建立包含眼睛、鼻子、嘴巴以及耳朵的面部特征子集,与标准图像比对,采用最大期望算法提取重叠度最高的面部特征,作为人脸识别的参照标准。采用两个向量的内积衡量人脸图像与参考值之间的相似度,计算存在和不存在局部遮挡的特征权重,建立置信区间,代入面部特征权重实施置信对比,完成人脸识别。实验结果表明,所提方法的识别精准度较高,在灰度变化、噪声以及局部遮挡的情况下均能够保证识别精度。

基于双平滑函数秩近似和群稀疏的高光谱图像恢复模型

高光谱图像(HSI)具有良好的光谱识别能力,被广泛地应用于各种领域。然而,HSI在成像过程中易受到混合噪声的污染,会严重削弱后续任务的准确性,如何高质量地恢复HSI是需要解决的首要问题。目前,基于低秩先验和全变分正则化结合的HSI去噪方法取得了较好的性能,但这些方法一方面忽略了高强度条纹噪声在空间结构和光谱分布上的特征,使得噪声无法完全去除,另一方面没有考虑HSI差分图像低秩子空间的信息,不能挖掘潜在的局部空间光滑结构。为此,提出了一种基于双平滑函数秩近似和群稀疏的HSI恢复模型。首先,利用双平滑函数秩近似模型探索干净HSI和条纹噪声的低秩结构,去除结构化条纹噪声等高强度混合噪声。其次,将基于E3DTV的群稀疏正则化融入双平滑函数秩近似模型中,充分挖掘HSI差分图像的稀疏先验信息,进一步提升图像在空间恢复和光谱特征保留方面的性能。最后,设计了交替方向乘子法(ADMM)求解所提出的BSRAGS模型。仿真和真实数据实验均表明,所提模型能够有效提高图像恢复质量。

基于RRT算法的移动机器人安全光滑路径生成

在多障碍物复杂工厂环境中,针对快速探索随机树算法(RRT)生成的路径存在冗余点、贴近障碍物且存在锯齿状转折的问题,改进得到了安全-光滑RRT(Safe-SmoothRRT)路径规划算法。首先,引入目标偏置策略;其次,该算法利用融合目标点引力思想的新节点扩展方式以及改进的近邻点度量策略以减少树的盲目扩展,提高生长的目标性;随后,引入节点安全约束,将安全节点加入树中;改进路径简化方法,剔除冗余点的同时兼顾了安全性;最后通过B样条局部平滑来改善路径的平滑性。在MATLAB仿真实验中分别与标准RRT算法、自适应目标偏向性RRT算法和改进RRT算法相比,在平均路径长度方面最大下降了7.1%,在平均有效节点数方面最大下降了64.1%,且所得路径始终与障碍物保持一定的安全距离,结果表明改进算法有效提升了路径的光滑性和安全性。

一种考虑关节跃度约束的实时刀具路径光顺算法

针对机器人加工过程中关节跃度超过约束的问题,提出一种考虑机器人关节跃度约束的刀具路径光顺算法。采用有限冲激响应滤波器插值算法对五轴刀具路径进行光顺处理,一步完成了路径光顺和速度规划。建立了关节与刀具之间的速度、加速度和跃度关系,通过调整时间常数保证了关节跃度不超过约束。考虑机器人关节加速度约束后,该算法计算出的时间常数更小。仿真和实验结果表明该算法能减小关节的跟踪误差,提高机器人的工作稳定性。

Numerical method for dynamics of multi-body systems with two-dimensional Coulomb dry friction and nonholonomic constraints

Based on the dynamical theory of multi-body systems with nonholonomic constraints and an algorithm for complementarity problems, a numerical method for the multi-body systems with two-dimensional Coulomb dry friction and nonholonomic constraints is presented. In particular, a wheeled multi-body system is considered. Here, the state transition of stick-slip between wheel and ground is transformed into a nonlinear complementarity problem （NCP）. An iterative algorithm for solving the NCP is then presented using an event-driven method. Dynamical equations of the multi-body system with holonomic and nonholonomic constraints are given using Routh equations and a con- straint stabilization method. Finally, an example is used to test the proposed numerical method. The results show some dynamical behaviors of the wheeled multi-body system and its constraint stabilization effects.

The Non-Smooth Problem of Wheel Motion

Planar motion of a non-deformable wheel under the action of non-ideal unilateral constraints is considered. The mathematical description of this phenomenon has a form of a non-smooth initial value problem. The non-smoothness of this problem means that its solution is determined by an absolutely continuous function having a discontinuous first derivative. For this reason, a collision problem describing abrupt changes of velocity has been formulated next to the equations of motion specifying the acceleration. The non-idealness of constraints means that the constraint reaction force includes also a component resulting from the friction between the wheel and the constraints. Differential equations specifying acceleration of the wheel making contact with the constraints and algebraic equations for determining the changes in the wheel’s velocity at the moment of collision have been formulated in the paper. The principal task in these formulations is to determine the reaction forces of the considered constraints. This task is specified by the relationships between acceleration and the constraint reaction force components. In the description of the collision, these relations refer to the post-collision velocities and reaction force impulses. For determining an approximate solution of the formulated wheel motion problem, an original numerical method and a computer program for wheel motion simulation have been developed. Selected results illustrating the changes in displacements and velocity have been presented.

面向电离层反演的参数平滑迭代层析方法

层析是电离层三维结构反演的重要技术.增加约束可缓解层析模型的病态问题.然而,迭代层析中,大量无射线穿过的网格易对施加的约束产生与真实意图相违背的约束结果(称负面约束).针对该问题,提出了一种参数平滑的迭代层析方法(iterative tomography method via parameters smoothing, ITPS).每次迭代中,先用Chapman函数及最小二乘法拟合及改正每一垂直剖面,获取关于Chapman函数参数的二维图像,然后用移动窗口法平滑各参数图像并以此改正各网格的电子密度.实验表明:ITPS方法可一定程度上减少负面约束并抑制层像的扰动.相对于MART和CMART算法,ITPS方法在垂直剖面、F2层临界频率(f_(o)F_(2))、F_(2)层峰值高度(h_(m)F_(2))、斜电子总含量(Slant Total Electron Content, STEC)及h_(m)F_(2)以上电子密度等方面均具有更佳的表现.相对于精度较高的CMART算法,ITPS方法在f_(o)F_(2)与h_(m)F_(2)的平均优化率分别为7.49%及6.60%,STEC的平均优化率为5.19%,h_(m)F_(2)以上电子密度的平均优化率为11.41%.

基于单色LED的日光光谱拟合

在照明领域中,LED已被证明是一种很有前途的人造光源,其光谱可以用来合成日光光谱。目前通过不同种类的LED合成日光光谱的关键问题是如何从LED数据库中选择LED型号和确定各型号LED的工作参数。针对这一问题,科学家们提出了通过各种数学模型来表示LED的光谱功率分布,如高斯函数、洛伦兹函数、logistic功率峰函数等。这些数学模型可以准确地反映出LED的非线性光谱特征,包括峰值波长、半高宽、红移等。但是通过数学光谱模型模拟的LED光谱与真实的LED光谱并不完全相同,通过数学光谱模型模拟的LED光谱曲线都是轴对称的。然而,真实的蓝色和绿色LED光谱曲线在短波长的一面更陡,在长波长的一面更缓,红色LED光谱曲线则相反。基于现有的多个LED的基础上提出了一个新的模拟日光光谱的方法。通过设计一个滤光片和调节各个现有的LED光源的辐射量权重(LED光源的辐射量权重为当前LED光源的辐射量与该LED的最大辐射量的百分比)来进行仿真实验,以此模拟出一个新的光源,使其光谱与日光光谱相接近。即将一个滤光片放置在多个LED的混合光源下,通过滤光片的光谱透射率和各个LED光源的辐射量权重来限制光源的光谱,使得修正后的多个LED混合光源的光谱尽可能的接近日光光源光谱。因工业要求需要对滤光片的光谱透射率曲线进行平滑约束。最后通过优化后的滤光片的光谱透射率求出各个LED光源的辐射量权重的最小二乘解,从而得到各个最优的LED光源的辐射量权重。通过仿真实验,最后该方法得到的模拟日光光源的相关色温(CCT)为6492 K,目标光源的相关色温为6503 K。模拟日光光源光谱与日光光谱的拟合优度(R 2)为0.9926。该方法对基于LED的日光光谱拟合的研究有重要的参考价值。

Android智能手机星间单差法伪距平滑算法研究

受到手机芯片和天线等影响,智能手机原始伪距观测值的噪声较大,同时受多路径影响严重,导致手机定位不准确、轨迹不平滑。针对上述问题,提出一种星间差载波相位平滑伪距、星间差多普勒平滑伪距和星间差速度约束自适应切换滤波定位算法,可充分发挥相位和多普勒观测值的作用,提高伪距观测值的精度并减少平滑伪距噪声。在静态和动态环境中分别对单频和双频智能手机定位性能进行分析。结果表明,双频定位效果优于单频,星间差速度约束自适应切换滤波定位算法最优。与抗差伪距法相比,静态环境下小米8在水平和高程方向上的精度分别提升1.49 m和1.38 m,华为P40分别提升2.11 m和1.53 m;动态环境下小米8在水平和高程方向上的精度分别提升1.75 m和1.25 m,华为P40分别提升2.58 m和1.85 m。

THE TOPOLOGICAL OPTIMIZATION FOR TRUSS STRUCTURES WITH STRESS CONSTRAINTS BASED ON THE EXIST-NULL COMBINED MODEL

A new exist-null combined model is proposed for the structural topology optimization. The model is applied to the topology optimization of the truss with stress constraints. Satisfactory computational result can be obtained with more rapid and more stable convergence as compared with the cross-sectional optimization. This work also shows that the presence of independent and continuous topological variable motivates the research of structural topology optimization.

基于优化人工势场法的无人机航迹规划

针对传统人工势场(traditional artificial potential field,TAPF)法在无人机航迹规划中存在的局部极小值、斥力过大、无效避障等问题,提出一种优化人工势场法。首先将障碍物斥力进行分解,避免了局部极小值情况;其次重构合力计算方式,避免无人机在障碍密集区域所受斥力过大;最后引入二次碰撞预测方法,减少无人机无效避障的同时保证航迹平滑。在考虑无人机物理约束条件下进行航迹规划实验。仿真结果表明,该方法相较于TAPF法,不仅缩短了规划航线长度,且在航迹平滑性上有明显提升。

基于关节电机性能约束机器人的运动轨迹平滑规划

为了使机器人运动过程平滑,采用机器人的逆动力学特性,通过约束机器人关节电机性能得到一种运动轨迹平滑规划方法。该方法中平滑运动轨迹使用五次样条曲线,以笛卡尔空间粗插补和关节空间精插补相结合,在考虑关节角速度约束和逆动力学得到的关节力矩约束下,建立角速度和关节力矩等关节电机性能与粗插补周期的关系式,便可通过修改粗插补周期得到满足关节电机性能约束的平滑运动轨迹,避免因优化平滑运动轨迹而带来的大量计算,更容易操作。通过与三次B样条曲线插补算法规划的轨迹相比,结果表明:此规划能大大提升机器人关节运动的平稳性,减小冲击和振动对机器人运动的影响。

用改进的光滑约束最小二乘正交分解法实现电阻率三维反演

对三维电阻率反演问题进行了深入研究,提供了一种利用地表观测数据实现三维反演的实用算法.该方法应用有限差分求正演解,并通过对粗糙度矩阵元素进行适当改进,使之适用于各种情况下粗糙度矩阵的求取,进而建立在模型的总粗糙度极小条件下的反演方程.对反演方程采用收敛速度快且稳定的最小二乘正交分解(LSQR)法进行迭代求解,在迭代求解过程中只需利用偏导数矩阵和其转置矩阵乘以一个向量的结果,回避了直接求偏导数矩阵的繁琐计算,节省了内存,加快了反演的计算速度.不同的计算实例表明上述方法是求解大规模三维电阻率反演问题的有效方法.

复杂约束限制下的梯级水电站群实时优化调度方法及调整策略

由于负荷预测偏差、径流预报不准确,如何快速、可靠地调整发电计划是梯级水电群实时调度中的难点和关键问题,主要体现在快速躲避机组限制区、兼顾电站日前计划和出力平稳性、最小化电站弃水等3个方面。针对上述问题,提出复杂约束限制下的梯级水电站群实时优化调度方法及调整策略,包括电站计算序位确定、电站组合确定、负荷偏差光滑处理、厂间负荷偏差分配和弃水调整策略。引入负荷偏差光滑技术,重构负荷偏差,保证厂间负荷偏差满足出力平稳性要求;在厂间负荷偏差分配过程中,协调日前计划,动态调整电站出力,避免落入机组限制区;采用弃水修正策略,沿约束边界调整电站出力,实现预泄腾库。所提方法和策略,能够充分考虑日前发电计划指导性,避免电站出力频繁波动,保证机组实时躲避限制区运行,避免不必要弃水发生。贵州电网乌江流域梯级水电站群厂网协调实际过程模拟分析表明,所提方法和策略能够快速可靠地满足实际工程要求,是一种是切实可行的方法和策略。