面向大规模集群的并行I/O用户层配置优化策略

影响应用I/O性能的关键因素主要有三个层次:包括应用的I/O接口实现、体系结构和文件系统组件的性能以及应用的I/O参数配置。从应用I/O配置优化的视角,分析了大规模集群并行I/O的配置调优空间,在此基础上,给出了一套大规模集群并行I/O性能特征测试分析方法。基于该方法,在某国产超级计算集群上开展了一系列I/O测试分析来刻画系统的I/O性能特征,进而指导并行应用程序的I/O配置优化。基于优化后的配置参数,在两类典型的并行I/O场景中,针对某类生产应用程序,8192进程下的重启动数据写操作时间下降了15%,4096核的程序作业加载时间从10 min缩短到了5 s。     

Experimental investigation on dynamic response and damage models of circular RC columns subjected to underwater explosions

Reinforced concrete (RC) columns are widely used as supporting structures for high-piled wharfs. The study of damage model of a RC column due to underwater explosion is a critical issue to assess the wharf’s antiknock security. In this study, the dynamic response and damage model of circular RC columns subjected to underwater explosions were investigated by means of scaled-down experiment models. Experiments were carried out in a 10.0 m diameter tank with the water depth of 2.25 m, under different explosive quantities (0.025 kg–1.6 kg), stand-off distances (0.0 m–7.0 m), and detonation depths (0.25 m–2.0 m). The shock wave load and dynamic response of experiment models were measured by configuring sensors of pressure, acceleration, strain, and displacement. Then, the load distribution characteristics, time history of test data, and damage models related to present conditions were obtained and discussed. Three damage models, including bending failure, bending-shear failure and punching failure, were identified. In addition, the experience model of shock wave loads on the surface of a RC column was proposed for engineering application.     

基于正态分布曲线的分段式变步长LMS算法

针对传统最小均方误差(Least Mean Square, LMS)自适应滤波算法由于步长固定,在解决稳态误差与收敛性之间的关系时,始终处于矛盾状态的问题,在对传统的固定步长LMS自适应滤波算法分析的基础上,根据变步长LMS自适应滤波算法的步长调整原则,通过构造步长因子与误差信号的非线性函数,提出了一种基于正态分布曲线的分段式变步长LMS自适应滤波算法,并分析了参数取值对算法性能的影响。针对实际信号处理过程中参考信号难以选取的问题,提出了一种基于分裂阵的参考信号选取方法。理论和海试数据分析结果表明:该算法的收敛速度和稳态误差明显优于固定步长的LMS自适应滤波算法和基于Sigmoid函数的变步长LMS自适应滤波算法。     

基于遗传算法的应急物流多设施选址模型研究

在现有研究的基础上,建立了应急物流多设施选址模型。在模型建立过程中除了考虑距离这一基本要素,还考虑了流量与安全性等道路特性,并将其转换为道路的综合权值,使模型更具现实意义。采用遗传算法求解,使该问题的求解过程得到简化。最后用实例对模型进行了验证分析,说明该模型的合理性。     

不确定多属性决策的单目标优化方法研究

运用神经网络的非线性映射和遗传算法的寻优特性,建立了不确定多属性决策的单目标优化模型。把属性值区间作为遗传算法染色体的搜索范围,用训练好的神经网络计算适应度。用不同的适应度函数来计算综合属性值区间数的下界和上界,然后对方案进行排序。算例结果表明,该方法是可行的。     

Fragment spatial distribution of prismatic casing under internal explosive loading

Non-cylindrical casings filled with explosives have undergone rapid development in warhead design and explosion control. The fragment spatial distribution of prismatic casings is more complex than that of traditional cylindrical casings. In this study, numerical and experimental investigations into the fragment spatial distribution of a prismatic casing were conducted. A new numerical method, which adds the Lagrangian marker points to the Eulerian grid, was proposed to track the multi-material interfaces and material dynamic fractures. Physical quantity mappings between the Lagrangian marker points and Eulerian grid were achieved by their topological relationship. Thereafter, the fragment spatial distributions of the prismatic casing with different fragment sizes, fragment shapes, and casing geometries were obtained using the numerical method. Moreover, fragment spatial distribution experiments were conducted on the prismatic casing with different fragment sizes and shapes, and the experimental data were compared with the numerical results. The effects of the fragment and casing geometry on the fragment spatial distributions were determined by analyzing the numerical results and experimental data. Finally, a formula including the casing geometry parameters was fitted to predict the fragment spatial distribution of the prismatic casing under internal explosive loading.     

面向大规模集群并行I/O的用户层配置优化策略研究

近年来，大规模并行场景下的I/O性能越来越受到应用科学家的关注。影响应用I/O性能的关键因素主要有三个层次：包括应用的I/O接口实现、体系结构和文件系统组件的性能以及应用的I/O参数配置。本文从应用I/O配置优化的视角，分析了大规模集群并行I/O的配置调优空间，在此基础上，给出了一套大规模集群并行I/O性能特征测试分析方法，基于该方法，在某国产超级计算集群上开展了一系列I/O测试分析来刻画系统的I/O性能特征，进而指导并行应用程序的I/O配置优化。基于优化后的配置参数，在两类典型的并行I/O场景中，针对某类生产应用程序，8192进程下的重启动数据写操作时间下降了15%，4096核的程序作业加载时间从10分钟缩短到了5s。本文提出的I/O配置调优空间及优化方法，可以推广应用到其它同类系统平台，对于大规模集群上的用户层并行I/O配置调优具有借鉴意义。     

单机攻击多目标逻辑的遗传优化求解

攻击逻辑的优化决策对于机载多目标攻击武器系统具有重要意义。本文建立了单机多目标攻击逻辑的解算模型 ,提出了解算要求 ,给出了采用遗传算法求解多目标攻击逻辑的应用方案。在此基础上 ,采用该方案进行大量的试探性仿真研究 ,对仿真结果进行了分析 ,并选择了合适的遗传算法参数。文章最后阐述了用遗传算法求解多目标攻击逻辑问题的发展前景