基于机器学习的动态分区并行文件系统性能优化

近年来,随着大数据、云计算技术的发展,应用系统越来越集中,规模亦越来越大,使得存 储系统的性能问题越来越突出。为应对其性能要求,并行文件系统得到了大量的应用。然而现有的并 行文件系统优化方法,大多只考虑应用系统或并行文件系统本身,较少考虑两者之间的协同。该文基 于应用系统在并行文件系统上的访问模式对存储系统的性能有显著影响这一特点,提出基于动态分区 的并行文件系统优化方法。首先,利用机器学习技术来分析挖掘各个性能影响因素和性能指标之间的 关系和规律,生成优化模型。其次,以优化模型为基础,辅助并行文件系统的参数调优工作。最后, 基于 Ceph 存储系统进行原型实现,并设计了三层架构应用系统进行了性能测试,最终达到优化并行 文件系统访问性能的目的。实验结果表明,所提出方法可以达到 85% 的预测优化准确率;在所提出模 型的辅助优化下,并行文件系统的吞吐量性能得到约 3.6 倍的提升。     

Multi-Criteria Query Optimization in the Presence of Result Size and Quality Tradeoffs

In this paper, we present novel multi-criteria query optimization techniques for performing query optimization in databases, such as multimedia and web databases, which rely on imperfect access mechanisms and top-k predicates. We present an optimization model that (1) takes into account different binding patterns associated with query predicates, (2) considers the variations in the expected query result sizes as a function of query execution plans, and (3) considers the expected result qualities of the execution orders. We address the complexity and the well-known NP-complete nature of the query optimization problem by adaptively reducing the granularity of the search space. For this purpose, unlike the data histograms which capture the data distribution, we propose opt-histograms that capture the distribution of sub-query-plan values over many optimization tasks.     

液压驱动六足机器人步行腿节段长度比例研究

针对液压驱动六足机器人三节段步行腿,建立了运动学模型,规划了足端运动轨迹,分析了腿长比例对步行速度、关节速度和加速度、足端工作空间及机体灵活度的影响.发现基节比例在0~0.1之间、大腿比例在0.4~0.5之间时,各项指标处于合理范围.根据0.09:0.455:0.455的腿长比例建立了腿部实物模型,搭建了腿部性能测试系统.试验结果表明,关节运动过程中未到达液压缸极限位置且无死点;关节最大速度与加速度满足性能需求;各关节最大速度复现偏差小于0.2?/s,最大加速度复现偏差小于0.7?/s2.验证了腿部节段长度分析结果的合理性与可行性.     

六边形对称分布六腿机器人的典型步态及其运动性能分析

为了便于在不同地理条件下合理地选择较优的步态,实现稳定高效的智能行走,本文针对一种六边形 对称分布的六腿机器人研究其不同步态的优劣．主要从行走能力、稳定性和能耗3 个角度对六边形对称结构的六腿 机器人在同样占空比下的3 种静态稳定周期步态进行了比较研究,此外还简要分析了其越障能力和穿越窄道的能 力．研究分析结果表明3 种步态（横向昆虫式摆动步态、哺乳动物式踢腿步态和混合步态）在不同条件下各有优劣： 横向昆虫式摆动步态在能耗和越障能力方面较其他两种步态有优势;而混合步态在稳定性上最具优势,其它能力处 于中间;哺乳动物式踢腿步态则可穿越窄道,步长上较昆虫摆动步态略好．本文的研究工作为六边形对称结构的六 腿机器人在未知复杂地貌环境下的智能行走提供了重要参考．     

复杂并行机调度问题基于分解的优化算法

针对纺织生产过程中广泛存在的带特殊工艺约束的大规模并行机调度问题，提出了一种基于分解的优化算法。首先将原调度问题分解为机台选择和工件排序两个子问题，然后针对机台选择子问题提出一种进化规划算法，并采用一种具有多项式时间复杂度的最优算法求解工件排序子问题，以得到问题特征信息（即每台机器对应拖期工件数的最小值），该问题特征信息用以指导进化规划算法的迭代过程。不同规模并行机调度问题的数值计算结果及实际制造企业应用效果表明，本文提出的算法是有效的。     

可扩展机器学习的并行与分布式优化算法综述

机器学习问题通常会转换成一个目标函数去求解,优化算法是求解目标函数中参数的重要工具.在大数据环境下,需要设计并行与分布式的优化算法,通过多核计算和分布式计算技术来加速训练过程.近年来,该领域涌现了大量研究工作,部分算法也在各机器学习平台得到广泛应用.本文针对梯度下降算法、二阶优化算法、邻近梯度算法、坐标下降算法、交替方向乘子算法五类最常见的优化方法展开研究,每一类算法分别从单机并行和分布式并行来分析相关研究成果,并从模型特性、输入数据特性、算法评价、并行计算模型等角度对每个算法进行详细对比.随后对有代表性的可扩展机器学习平台中优化算法的实现和应用情况进行对比分析.同时对本文中介绍的所有优化算法进行多层次分类,方便用户根据目标函数类型选择合适的优化算法,也可以通过该多层次分类图交叉探索如何将优化算法应用到新的目标函数类型.最后分析了现有优化算法存在的问题,提出可能的解决思路,并对未来研究方向进行展望.     

并联机床数控系统运动学变换的研究

并联机床由于完全摒弃了传统机床固定导轨的刀具导向方式,其数控系统中必然需要添加运动学变换模块,以实现加工代码中的直角坐标与其可控制的关节空间坐标之间的转换.本文介绍了3-TPS并联机床数控系统中运动学变换方程组的建立和求解,详细分析了五坐标加工时运动学变换的实现,并在实际平台中仿真测试了运动学模块的性能.     

仿生六足机器人多电机控制系统的研究与设计

以电机专用控制芯片F2812 DSP为核心构建了仿生六足机器人多电机控制系统.设计和编制了相应的硬件电路和软件流程,使PWM周期达到20μs左右,将机器人控制系统的整体性能大大提高;仿真实验结果证明,该控制系统运行稳定、性能可靠,为仿生六足机器人技术的进一步完善奠定了基础;所探讨的技术方法和设计思路还可为其他类型机器人的开发和研制提供借鉴和参考。     

The Optimal Design of a Planar Parallel Platform for Prescribed Machining Tasks

In this paper a general methodology is presented for determining theoptimum design of a planar parallel platform to be used in machining.The method, based on a mathematical optimization approach, is used tofind a platform design and placement such that firstly, the execution ofa prescribed task path is feasible, and secondly, the actuator forcesrequired to execute the prescribed task are minimized. The applicationof the method is illustrated for two prescribed tasks, five designvariables and a number of geometrical inequality constraints such asactuator length limits. The method succeeds in finding locally optimumand feasible platform designs for which the required task lies insidethe workspace. Two optimization algorithms are implemented and theirrespective results are compared. The first algorithm is a robust andreliable trajectory algorithm, LFOPC, which is however expensive interms of the number of required function evaluations. As the simulationsperformed here in evaluating the objective and constraint functions maybe computationally intensive, an approximation method, Dynamic-Q, isalso used to find the optimum design with greater efficiency. Theeffectiveness of this approximation approach is evaluated.     

A New Approach to the Architecture Optimization of a General 3-PUU Translational Parallel Manipulator

This paper presents a new approach to the architecture optimization of a general 3-PUU translational parallel manipulator (TPM) based on the performance of a weighted sum of global dexterity index and a new performance index-space utility ratio (SUR). Both the inverse kinematics and forward kinematics solutions are derived in closed form, and the Jacobian matrix is derived analytically. The manipulator workspace is generated by a numerical searching method with the physical constraints taken into consideration. Simulation results illustrate clearly the necessity to introduce a mixed performance index using space utility ratio for architectural optimization of the manipulator, and the optimization procedure is carried out with the goal of reaching a compromise between the two indices. The analytical results are helpful in designing a general 3-PUU TPM, and the proposed design methodology can also be applied to architectural optimization for other types of parallel manipulators.     

Kinematics,dynamics and dimensional synthesis of a novel 2-DoF translational manipulator

In the past few years, parallel manipulators have become increasingly popular in industry, especially, in the field of machine tools. In this paper, a novel 2-degree-of-freedom (DoF) parallel manipulator, which has two translational DoFs, is proposed. It is characterized by the fact that the output of the manipulator is two planar DoFs of a rigid body, while its orientation remains constant. The inverse and forward kinematics can be described in closed form. The velocity equation, singularity, and workspace of the manipulator are presented. In addition the inverse dynamics problem of the device is investigated employing the Lagrange multipliers approach. The dimensional synthesis based on the workspace and conditioning indices is presented. The proposed manipulator can be applied to the field of machine tools or used as the mobile base for a spatial manipulator. The results of the paper are very useful for the design and application of the new manipulator.     

立式混合机温控系统分析与设计

按照大型立式混合机生产工艺与设备要求，介绍大型立式混合机在进行物料混合生产工艺时的温度自动控制系统的工作特点与设计方法，设计出了满足生产实际要求的结构简单性能优良的温度自动控制系统。     

Design-by-Optimization and Control of Redundantly Actuated Parallel Kinematics Sliding Star

The paper deals with the design and control of an example of redundantly actuated parallel kinematic structure that can be a machine tool. The principle of redundant actuation brings parallel kinematic structures which do not have singular positions in workspace and which has increased dynamic, stiffness and accuracy properties. There is proposed a parallel kinematic structure called Sliding Star that has promising dynamic and stiffness properties. The conceptual design-by-optimization of this structure is briefly described. The redundantly actuated parallel kinematics have control problem due to mutual fighting of redundant drives. There is described the solution of this problem. Based on the investigated redundantly actuated parallel kinematics there has been built a laboratory prototype. The experimental results from the control of this prototype are briefly presented.     

基于粒子群优化的支持向量机在机械模式分类中的应用

机械故障诊断本质上是一个模式分类问题.支持向量机由于解决分类问题有着较好的表现,得到了日益广泛的应用.针对支持向量机的参数对分类性能的影响,采用粒子群算法对支持向量机的惩罚因子和径向基核函数进行优化,使支持向量机的分类性能最优,并将其应用于实例,得到了较好的分类正确率.     

模糊控制器的优化设计方法

本文提出了一种模糊控制器的优设计方法，讨论了直接优化去模糊过程的控制规则，并对基于稳定和不稳定模型进行了仿真研究。     

Design Optimization of Multibody Systems by Sequential Approximation

Design optimization of multibody systems is usually established by a direct coupling of multibody system analysis and mathematical programming algorithms. However, a direct coupling is hindered by the transient and computationally complex behavior of many multibody systems. In structural optimization often approximation concepts are used instead to interface numerical analysis and optimization. This paper shows that such an approach is valuable for the optimization of multibody systems as well. A design optimization tool has been developed for multibody systems that generates a sequence of approximate optimization problems. The approach is illustrated by three examples: an impact absorber, a slider-crank mechanism, and a stress-constrained four-bar mechanism. Furthermore, the consequences for an accurate and efficient accompanying design sensitivity analysis are discussed.     

基于角度优化的鲁棒极端学习机算法

极端学习机因其学习速度快、泛化性能强等优点,在当今模式识别领域中已经成为了主流的研究方向;但是,由于该算法稳定性差,往往易受数据集中噪声的干扰,在实际应用中导致得到的分类效果不是很显著;因此,为了提高极端学习机分类的准确性,针对数据集样本中带有噪声和离群点问题,提出了一种基于角度优化的鲁棒极端学习机算法;该方法利用鲁棒激活函数角度优化的原则,首先降低了离群点对分类算法的影响,从而保持数据样本的全局结构信息,达到更好的去噪效果;其次,有效的避免隐层节点输出矩阵求解不准的问题,进一步增强极端学习机的泛化性能;通过应用在普遍图像数据库上的实验结果表明,这种提出的算法与其他算法相比具有更强的鲁棒性和较高的识别率。     

一种四自由度并联机构的误差分析及其标定补偿

针对一种四自由度并联机构进行了运动学分析,在此基础上以摄动法建立了其误差模型,明确了各误差源对末端位姿的影响．在对并联机构的标定技术进行简单介绍后,说明了对该机构误差模型中的机构参数进行标定的两种方法,介绍了标定装置、标定算法及标定过程．最后采用Matlab对基于逆解的标定方法进行了仿真,并对仿真结果进行了分析．     

基于多蚁群算法的支持向量回归机参数选择方法

核函数是支持向量回归机的重要部分,每种核函数都有其优势和不足。本文基于支持向量机回归机模型相关参数的选取原则,给出了一种具有混合核函数的支持向量机,以基于网格搜索的多蚁群算法为基础,给出了此类混合核函数支持向量回归机参数优化的一种新方法。该方法以最小化交叉验证误差为目标,对包括混合比例和各类核函数的参数在内的5个参数进行优化。仿真结果表明,与遗传算法相比,本方法在参数优化方面有良好的性能,建立的预测模型精度较高。     

Research on Optimization Strategy of Missile Preliminary Design

Two system optimization architectures are proposed for missile system preliminary design, taking into account aerodynamics, weights and sizing, propulsion and trajectory. Approximation methods are investigated in order to reduce problem dimensionality and to improve the efficiency of optimization process.