利用Microsoft Virtual Server迁移遗留的应用程序从物理机器迁移到虚拟环镜

揭密如何利用Microsoft Virtual Server的虚拟机把老式应用程序从物理机器移植到虚拟环境中。

基于物理启发机器学习的属性散射中心提取方法

基于参数化散射中心模型进行参数估计是实现合成孔径雷达高级信息获取(SAR AIR)技术的基本思路之一,传统的属性散射中心(ASC)参数估计算法往往具有计算速度慢、算法复杂度高、对参数初值要求高等问题。对此,该文提出一个新的基于无监督学习的端到端框架用于从SAR图像反演ASC参数。首先,利用自编码式网络结构有效提取目标图像特征,缓解由于优化空间复杂非凸导致的直接求解困难,解决初值敏感问题;其次,通过嵌入ASC模型作为物理解码器以将编码器输出约束为正确的ASC参数;最后,通过端到端的模型架构进行学习和推理,达到降低算法复杂度及提高估计速度的目的。通过在仿真和实测数据上进行测试,实验结果表明在0.15 m分辨率测试集SAR图像上取得低于0.1 m的估计误差,反演单个散射中心平均耗时0.06 s,验证了该文所提方法的有效性、高效性与鲁棒性。

面向电力系统智能分析的机器学习可解释性方法研究(二):电网稳定分析的物理内嵌式机器学习

将知识融入机器学习模型是提升算法可解释性与计算性能的重要途径之一。针对电力系统运行的稳定分析问题,提出一种新的物理内嵌式机器学习框架与方法,将描述故障动态过程的微分-代数方程作为先验知识,引导神经网络模型训练。相比于完全依赖数据的通用机器学习方法,物理内嵌式机器学习直接模拟物理过程,通过数据背后所蕴含的物理方程来约束机器学习决策空间,并输出故障后的动态曲线,结果物理含义与可解释性更强。同时,物理内嵌模式也大幅降低了模型训练对海量数据的依赖,为小样本学习及以及模型向真实系统迁移应用过程中的参数辨识提供一种新的思路。

基于JEC-FDTD等效循环神经网络的电磁建模和等离子体参数反演

磁化等离子体中的电磁波传播是重要的研究课题,针对特定场景下的电磁等离子耦合问题,进行有效且准确的方程建模与参数求解具有极强的研究价值和挑战性,这是探究电磁波与等离子体复杂非线性相互作用机制的关键。文中设计了一种可用于电磁等离子体正逆向建模的循环神经网络(recurrent neural network,RNN),该网络正向传播过程等价于任意磁倾角情况下的电流密度卷积时域有限差分(current density convolution finite-difference time-domain,JEC-FDTD)方法,因此可以求解给定的电磁建模问题,并易于大规模并行计算。通过构建前向可微模拟过程,JEC-FDTD方法可以使用自动微分技术准确且高效地计算梯度,然后通过训练网络来解决反问题。因此,该方法可以有效利用观测到的时域散射场信号反演重要的等离子体参数。JEC-FDTD方法和RNN相结合,形成了较强的协同效应,使得模型具有可解释性和高效的计算效率,受益于深度学习提供的优化策略和专用硬件支持,可以适用于不同仿真场景下的电磁建模和等离子体参数反演。

融合相似性原理的涡轮叶型流场预测方法研究

计算流体力学(CFD)方法是涡轮叶片等设计阶段性能评估的重要手段.然而,基于CFD的数值仿真方法通常比较耗时,难以满足涡轮叶型设计阶段快速迭代的需求.为实现快速性能评估并克服纯数据驱动预测模型泛化能力不足的问题,受到物理增强的机器学习思路的启发,将相似性原理与深度学习模型相结合,提出了一种泛化能力强的涡轮叶型流场预测新方法.以涡轮叶片表面等熵马赫数分布预测为例,提出采用相似性原理对叶型几何变量和气动参数进行归一化,进而在归一化参数空间构建训练样本集与深度学习预测模型,由此建立统一的流场预测模型,对几何尺寸、边界条件差异较大的叶型气动性能进行评估.在完成模型训练后,对归一化条件下不同工况/不同形状叶型的流场、真实环境下不同工况/不同尺寸叶型的流场以及GE-E3低压涡轮不同截面叶型的流场进行预测,结果表明预测结果的分布曲线与CFD评估结果吻合良好,平均相对误差在1.0%左右,由此验证了所提出的融合相似性原理的流场预测模型的精度与泛化能力.

感知物理先验的机器学习及其在地理空间智能中的研究前景

许多复杂的物理现象和动态系统尚未为科学家所完全理解和解释,因此不能构建确定的数学方程来描述,不能直接使用紧凑的物理机理模型来进行分析和预测。随着观测数据的日益丰富,数据驱动的机器学习方法可以较好地描述复杂非线性现象,但是纯数据驱动模型在表征能力、可解释、泛化能力、样本利用效率方面还存在很多不足。常规机器学习方法在地学领域的应用还面临时空异质性、样本稀疏等带来的挑战。近年来感知物理先验的机器学习方法可以在物理原理不太明确的情况下更有效地利用观测数据描述和分析动态系统,受到了广泛关注,并在物理、计算机、生物、医学和地学等领域得到了一定的应用。近年来人工智能和机器学习技术已经大量应用于地理学尤其是地理信息和遥感领域,受到地理学者广泛重视,被称为地理空间智能,并已成为一个热门的研究方向。感知物理先验的机器学习方法融合了模型驱动和数据驱动思想,为地理空间智能研究带来新的研究范式,促进各种复杂地理现象的精细描述和预测。本文首先分别从物理先验的表达形式和如何在机器学习中集成物理先验两方面概述了该领域的进展。物理先验是在机器学习建模之前已经存在、独立于机器学习方法的知识。本文从增广的数据和定制特征、物理定律和约束规则、支配方程、几何特性等方面总结物理先验知识的表达形式。主要从机器模型约束建模、辅助任务设计和模型的训练推理角度总结如何在机器学习模型中有效集成各种物理先验。本文基于以上的综述框架,结合地学机器学习和地理空间智能的发展和前沿问题,探讨了地理时空先验与其他物理先验的关系,简要总结和分析了目前感知时空先验的地理空间智能方法研究案例,探讨了时空先验表征以及集成时空先验地理空间智能的未来研究规划和应用前景。随着感知物理先验的机器学习方法研究的快速发展,我们相信感知时空先验的地理空间智能研究未来将逐步构建起跨多时空尺度的通用地理表征、分析、预测和解释框架,不仅能更好地解决地理信息科学的传统问题,还将鼓励地理学者与其他相关学科一起建立交叉研究的前沿机会,探索解决人类未来面临的共同挑战。

基于Markov过程的IaaS系统可用性建模与分析方法

针对现有基础设施即服务(IaaS)可用性模型难以计算存在多个可用物理机器(PM)概率的问题,提出一种基于Markov过程的IaaS可用性分析方法。首先,将计算资源划分为hot PM、warm PM和cold PM三类;然后,结合资源分配过程的相应阶段对可用性影响进行建模,分别生成对应的三种分配子模型,子模型之间通过不同种类计算资源的转换关系相互协作,构建系统整体模型;其次,基于Markov过程建立方程组以对可用性模型进行求解;最后,结合实例对分析模型进行验证,并对PM变迁速率等关键影响因素进行了分析。实验结果表明,增加PM尤其是cold PM的数量有助于提升IaaS的可用性。所提方法可以用于评估IaaS存在一个或多个可用PM的概率。

基于虚拟化技术实现MPICH2的高可用管理

为实现MPICH2的高可用,将MPICH2计算平台构建在虚拟机环境中,使得参与MPICH2运算的各个结点均是虚拟机,而物理机器并不直接作为结点参与运算;当物理机器需要进行维护、升级或负载均衡时,通过在线迁移的方式将其上运行的虚拟机迁移至另一个物理结点上,然后在新的结点上重新启动虚拟机,恢复其上运行的MPICH2计算任务,从而避免了终止整个计算任务的需要。实验结果表明,虚拟机在线迁移之后,整个MPICH2计算任务可以正确恢复;同时,比较了虚拟机和物理机器执行MPICH2计算任务的效率,表明了以虚拟机取代物理机器构建MPICH2计算平台在性能上是可行的。

Semi-physical simulation of AUV pipeline tracking

Before the task of autonomous underwater vehicle(AUV) was implemented actually,its semi-physical simulation system of pipeline tracking had been designed.This semi-physical simulation system was used to test the software logic,hardware architecture,data interface and reliability of the control system.To implement this system,the whole system plan,including interface computer and the methods of pipeline tracking,was described.Compared to numerical simulation,the semi-physical simulation was used to test the real software and hardware more veritably.In the semi-physical simulation system,tracking experiments of both straight lines and polygonal lines were carried out,considering the influence of ocean current and the situation of buried pipeline.The experimental results indicate that the AUV can do pipeline tracking task,when angles of pipeline are 15°,30°,45° and 60°.In the ocean current of 2 knots,AUV could track buried pipeline.

Research on Robotics Application in Fruit Harvesting

The study was aimed to identify the problems and also to suggest some solutions for robotic systems applications in harvesting of kiwi and apples. The results obtained show that physicomechanical properties of the harvested fruits had an effect on the use of robotic systems. Parameters such as weight, height, width, thickness and surface area were identified as the main effect in robotic fruit harvesting. Image processing techniques also had direct effect on robotic systems operation which therefore requires careful selection procedure to achieve results accuracy. It was observed that the C# programming language used in robotic systems fruit harvesting should be parallel with the image processing to ensure accurate conversion of kinematic calculation, system input parameters and constant values. The use of hobby-type parts of a robot in the prototype study showed reliable results. However, to ensure functionality of the robotic systems application, industrial robots and servo engines should be used. It was also observed that the use of a gripper for picking the fruits from the branch must The engines that make the system work should be strong and the be made considering the physicomechanical properties of the fruit. cutting system should also be appropriate.

Estimation of an Object＇s Physical Parameter by Force Sensors of a Dual-arm Robot

We are developing a nursing-care robot for physical care tasks. The concept of this robot is to promote the cared persons by the robot to activate their own motion ability as long as possible. This may lead to the improvement of the cared person＇s movement volition and movement abilities. In order to realize safe and human friendly robot care tasks, full body manipulation is an important technology, for which it is necessary to estimate the subject＇s center of gravity from the contact positions and forces with the robot＇s two arms. In this paper, we estimate the center of gravity of object based on the contact point and the contact force estimated by force sensor on both robot arms. The position of gravity center is important to realize care tasks stably. We performed experiments and simulations for the single point contact and dual points contact cases using a cylindrical object. As a result, it is found that although some errors were recognized in the experiments compared with the simulations, the relations between the contact positions and such errors were observed. Such experimental error mainly comes from the difference of shape between the real robot and the model of the robot in simulation.

Modeling of Welding Process by Robotic Vision

Modeling of welding process by robotic vision is basically a theoretical problem, means mainly on physical problem, and also technological problem. To obtain a good model of welding process by robotic vision, theoretical researches are required but also constant experimental researches of several welding processes. Until today researches of welding processes has been based on empirical and detailed experimentation. This paper presents welding process by robotic and automation points of view with application of new technologies. Welding robotic system has been designed with possibility to control and inspect this process. Parameters that should be controlled during the process have been identified to reach desired quality. Figure of control system of welding process by robotic vision is given in this paper.

From Digital Analogs Through Recursive Machines to Quantum Computer

The report examines the evolution of computers from digital analogs through non-yon Neumann machines to quantum computers, which are also digital analogs. In the 60 years of digital analogs successfully developed at the Institute of Electromechanics of the USSR in Leningrad. An important stage in the development of non-classical multiprocessor machine performance and reliability has been the development of recursive machines, which was carried out at the Institute of Cybernetics led V.M.Glushkov and the Leningrad Institute of Aviation Instrumentation. The general approach to the synthesis is carried out through linguo- combinatorial modeling with structured uncertainty.

土木工程智能科学与技术研究现状及展望

人工智能为土木工程学科和行业带来了新的发展机遇,土木工程智能科学技术的发展和创新与智能材料、3D打印、大数据、物联网、智能计算、机器人、元宇宙等数学、计算机科学和新兴产业技术进行深度融合,将深刻变革土木工程全寿期,包括城市规划、建筑设计、结构设计、工程建造、服役运维、防灾减灾等阶段。为此,从智能材料、智能设计、智能建造、智能运维、智能防灾等多个维度,总结当前土木工程智能科学与技术的研究现状,提出物理机器学习(基于数据范式与物理定律有机融合)的土木工程智能科学与技术特征;同时,探讨虚拟科学家在创造崭新土木工程智能科学与技术的前景。最后,展望了土木工程智能科学与技术发展目标,即从基于现有知识体系建立纯数据驱动和物理机器学习土木工程智能理论,到发展出具有自学习能力的智能体及其理论体系,最终创造出具有自我进化能力的虚拟科学家,发现新的数学方法与物理定律等。

Seeing permeability from images： fast prediction with convolutional neural networks

Fast prediction of permeability directly from images enabled by image recognition neural networks is a novel pore-scale modeling method that has a great potential. This article presents a framework that includes (1) generation of porous media samples,(2) computation of permeability via fluid dynamics simulations,(3) training of convolutional neural networks (CNN) with simulated data, and (4) validations against simulations. Comparison of machine learning results and the ground truths suggests excellent predictive performance across a wide range of porosities and pore geometries, especially for those with dilated pores. Owning to such heterogeneity, the permeability cannot be estimated using the conventional Kozeny–Carman approach. Computational time was reduced by several orders of magnitude compared to fluid dynamic simulations. We found that, by including physical parameters that are known to affect permeability into the neural network, the physics-informed CNN generated better results than regular CNN. However, improvements vary with implemented heterogeneity.

Levenberg-Marquardt based artificial physics method for mobile robot oscillation alleviation

This paper proposed a modified artificial physics(AP)method to solve the autonomous navigation problem for mobile robots in complex environments.The basic AP method tends to cause oscillations in the presence of obstacles and in narrow passages,which can result in time consumption.To alleviate oscillation,we modified the AP method using the Levenbery-Marquardt(LM)algorithm.In the modified AP method,we altered the original directions of AP forces to the Newton direction,and adjust the parameter by the LM algorithm.A series of comparative experimental results show that the modified AP method can achieve smoother trajectories with less time consumption.This demonstrates the feasibility and effectiveness of our proposed approach.

Self-assembly control and experiments in swarm modular robots

This paper presents a self-assembly control strategy for the swarm modular robots. Simulated and physical experiments are conducted based on the Sambot platform, which is a novel self-assembly modular robot having the characteristics of both the chain-type and the mobile self-reconfigurable robots. Multiple Sambots can autonomously move and connect with one another through self-assembly to form robotic organisms. The configuration connection state table is used to describe the configuration of the robotic structure. A directional self-assembly control model is proposed to perform the self-assembly experiments. The self-assembly process begins with one Sambot as the seed, and then the Docking Sambots use a behavior-based controller to achieve connection with the seed Sambot. The controller is independent of the target configuration. The seed and connected Sambots execute a configuration comparison algorithm to control the growth of the robotic structure. Furthermore, the simul- taneous self-assembly of multiple Sambots is discussed. For multiple configurations, self-assembly experiments are conducted in simulation platform and physical platform of Sambot. The experimental results verify the effectiveness and scalability of the self-assembly algorithms.