超燃冲压发动机燃烧室构型优化的试验研究

在直连式超燃冲压发动机试验系统上,通过调节超燃冲压发动机燃烧室壁面扩张角和燃料喷注位置,对燃烧室构型优化进行了试验研究。为了提高试验效率,燃烧室形面调节采用正交试验设计方法进行组织,每个形面进行5种喷注位置的试验,每次试验通过文氏管调节3个当量比的燃料流量。利用试验数据构造燃烧室性能关于构型参数的响应面模型,可用于燃烧室构型优化。通过两次渐进优化获得了性能更优的燃烧室构型,并根据试验数据分析了各构型参数对燃烧室性能的影响,结果表明:优化构型燃烧室的推力增益比基准构型增大了10.4%;燃烧室性能受各构型参数的强烈耦合影响。     

高性能RTI自适应通信机制研究

为了提高HLA仿真系统在高性能计算平台上的运行性能,针对高性能计算平台的高性能层次式通信环境,提出了一种共享内存/IB/以太网自适应RTI通信机制-SACM机制,该机制能够根据通信双方所在节点的网络拓扑信息,自适应建立延迟最小的通信连接,实现共享内存、IB等高速互连及以太网互连的自适应选择.同时在该通信机制中,针对节点内高速共享内存通信需求,提出了一种面向多盟员互连的共享内存通信算法,有效提高了节点内盟员通信效率.对象属性值更新延迟测试结果表明,该通信机制可以有效提高RTI在高性能计算平台的通信性能.     

雷诺数对沟槽减阻特性影响的数值分析

采用雷诺平均N-S方程和RNGk-ε湍流模型计算V型沟槽面的湍流边界层流动和黏性阻力,通过改变来流速度大小和沟槽面布置位置,研究了雷诺数对沟槽减阻特性的影响规律。计算结果表明,来流速度对沟槽减阻率的影响很大,对于一种尺度的V型沟槽,存在着一个具有较好减阻效果的来流速度范围,最大减阻率可达8.6%;沟槽面在沿来流方向上的布置位置对其减阻效果的影响则非常小。     

I-DEAS的三维几何建模

Ｉ－ＤＥＡＳ几何建模有两种方法。一是通过对基本体素进行布尔集合运算而构成复杂表体,称作体素拼接法;一是通过面扫描及点、线和面之间的运算操作构成三维实体,称作边界表示法。本文还介绍了Ｉ－ＤＥＡＳ几何建模的数据结构。     

基于混合挖掘的军事数据分类方法

基于军事信息系统的体系作战指挥,必须依托准确可靠的数据支撑。为了研究数据挖掘方法在军事数据上的学习性能,采用支持向量机方法和粗糙集对坦克连的作战数据进行了仿真实验。试验结果表明,支持向量机方法能够预测坦克连的命中率为100%,结合粗糙集方法,能够进一步提高支持向量机方法的学习性能,缩短学习时间。     

FlatLFS:一种面向海量小文件处理优化的轻量级文件系统

海量小文件的高效存储和访问是当前分布式文件系统面临的主要挑战之一。以GFS和HDFS为代表的分布式文件系统大多面向海量大文件的高效存储和访问设计,缺乏小文件处理的针对性优化,导致访问海量小文件时效率低下。针对分布式文件系统中海量小文件访问时的数据服务器优化问题,提出了一种采用扁平式数据存储方法的轻量级文件系统FlatLFS,取代传统文件系统对上层分布式文件系统提供数据存储和访问支持,提高了数据服务器处理小数据块时的I/O性能,从而提升了整个分布式文件系统的性能。实验表明,当数据块大小设定为1M时,FlatLFS的随机读性能分别比ext3、ext4、reiserfs高135%、112%和122%。     

基于SIMD处理器的全定制多粒度矩阵寄存器文件

在SIMD处理器上映射矩阵运算时会带来大量的数据重排操作从而降低系统性能。本文提出定制化的多粒度矩阵寄存器文件(MMRF)以消除数据重排操作。MMRF支持多粒度的并行行访问和列访问,从而提升矩阵运算的性能。MMRF可以被动态配置为不同的并行访问模式,在不同模式下一个或多个子矩阵可以被并行处理。实验结果显示,同传统的向量寄存器文件(VRF)和矩阵寄存器文件(MRF)相比,MMRF可分别带来2.21倍和1.6倍的平均性能提升,面积分别增加14.3%和3.7%,功耗分别增加14.6%和2.2%。同TMS320C64x+处理器相比,基于SIMD技术的FT-Matrix处理器在引入MMRF后可以得到5.65倍到7.71倍的性能提升。通过层次化的全定制设计技术,MMRF的面积和关键路径分别减少17.9%和39.1%。     

多核数字信号处理卷积算法并行优化

针对国防科技大学自主研发的异构多核数字信号处理(digital signal processing, DSP)芯片的特征以及卷积算法自身特点,提出了一种面向多核DSP架构的高性能多核并行卷积实现方案。针对1×1卷积提出了特征图级多核并行方案;针对卷积核大于1的卷积提出了窗口级多核并行优化设计,同时提出了逐元素向量化计算的核内并行优化实现。实验结果表明,所提并行优化方法实现单核计算效率最高能达到64.95%,在带宽受限情况下,多核并行扩展效率可达到48.36%～88.52%,在典型网络ResNet50上的执行性能与E5-2640 CPU相比,获得了5.39倍性能加速。     

基于动网格的翼型设计优化

针对传统基于代理模型的翼型优化的缺陷,提出采用基于动网格的翼型优化;针对基本Hicks-Henne方法后缘不光滑情况进行了改进;采用自编程序实现动网格生成,通过集成商业软件Pointwise和CFD计算软件Fluent完成边界条件生成和流场解算,并实现上述程序和软件间的数据交互和自动化,整个优化流程在iSIGHT平台下执行,优化算例表明基于动网格的翼型优化方法能大幅提高翼型性能,节省大量的重复性操作,增加优化结果的鲁棒性和可信度,是翼型优化中的一种有效方法。     

长向量处理器高效RNN推理方法

模型深度的不断增加和处理序列长度的不一致对循环神经网络在不同处理器上的性能优化提出巨大挑战。针对自主研制的长向量处理器FT-M7032,实现了一个高效的循环神经网络加速引擎。该引擎采用行优先矩阵向量乘算法和数据感知的多核并行方式,提高矩阵向量乘的计算效率;采用两级内核融合优化方法降低临时数据传输的开销;采用手写汇编优化多种算子,进一步挖掘长向量处理器的性能潜力。实验表明,长向量处理器循环神经网络推理引擎可获得较高性能,相较于多核ARM CPU以及Intel Golden CPU,类循环神经网络模型长短记忆网络可获得最高62.68倍和3.12倍的性能加速。     

面向图计算应用的处理器访存通路优化设计与实现

针对图计算应用的访存特点,提出并实现一种支持高并发、乱序和异步访存的高并发访存模块(High Concurrency and high Performance Fetcher, HCPF)。通过软-硬件协同的设计方法,HCPF可同时处理192条共8种类型的内存访问请求,且访存粒度可由用户定义,满足图计算应用对海量低延迟细粒度数据访问的需求。同时,HCPF扩展了基于内存语义的跨计算节点定制互连技术,支持远程内存的细粒度直接访问,为后续实现分布式图计算框架提供技术基础。结合上述两个核心研究内容,基于流水线RISC-V处理器核,设计并实现了可支持HCPF的RISC-V片上系统(System-on-Chip,SoC)架构,搭建基于FPGA的原型验证平台,并使用自研测试程序对HCPF进行初步性能评测。实验结果表明,HCPF相比原有访存通路,最高可将基于数组和随机地址的两种随机内存访问性能分别提升至3.5倍和2.7倍。远程内存直接访问4 Byte数据的延时仅为1.63μs。     

The relationship between decision variables and penalty cost parameter in (Q,R) inventory models

This paper is concerned with the optimum decision variables found using order quantity, reorder point (Q, R) inventory models. It examines whether the optimum variables (Q* and R*) are necessarily monotonic functions of the backorder cost parameter (or equivalently of the performance objective). For a general class of models it is proved that R* must increase as the performance objective is raised, and an inequality condition is derived which governs how Q* will change. Probability distributions of lead time demand are cited or found for which Q* increases, Q* decreases, and Q* is independent of increases in performance objectives or backorder cost parameter.     

图论中独立支配集的最佳求解算法研究

通过对图论中独立集和支配集的深入研究,提出了独立支配集的概念,论证了独立支配集同极大独立集及极小支配集之间的内在联系,并在此基础上给出了独立支配集的最佳求解算法,从而圆满地解决了图论中独立集及支配集的求解问题,对图的着色及匹配等问题的研究均有相当重要的借鉴意义。     

用于减少远程Cache访问延迟的最后一次写访问预测方法

为减少远程Cache访问延迟,提高共享存储系统的性能,提出了一种新的基于程序内在写突发特性的最后一次写访问预测方法,并对一个具体的目录协议进行了改造,以支持该预测方法。通过预测Cache块的最后一次写访问并提前对其进行降级,处理器能直接从主存中读取数据,从而减少了远程Cache访问所需的一个网络跳步数。与当前基于指令的预测方法相比,该方法能极大减少存储开销。基准测试程序的评测结果表明,该方法能获得83.1%的预测准确率,并且能提高8.57%的程序执行性能,同时与基于指令的预测方法相比,该方法能分别减少历史踪迹表69%的存储开销和签名表36%的存储开销。     

Iterative coloring extension of a maximum clique

In this paper we present an improved branch and bound algorithm for the vertex coloring problem. The idea is to try to extend the coloring of a maximum clique to its adjacent vertices. If this succeeds, its successive neighbors are considered; in case of failure (i.e., in the case the initial colors are not sufficient), working on the subgraph induced by the maximum clique and its neighborhood, the lower bound is improved by seeking for an optimal coloring of this subgraph by branch and bound. The process is repeated iteratively until the whole graph is examined. The iterative scheme exploits a further lower bound obtained by integrating a simple algorithm into the maximum clique search, and a new method to compute upper bounds on subgraphs. Furthermore, a new branching rule and a method for the selection of the initial maximum clique are presented. Extensive computational results and comparisons with existing exact coloring algorithms on random graphs and benchmarks are given. © 2001 John Wiley & Sons, Inc. Naval Research Logistic 48: 518–550, 2001     

An (s,S) model for inventory with exponential lifetimes and renewal demands

Inventory control of products with finite lifetimes is important in many modern business organizations. It has been an important and difficult research subject. Here, we study the (s, S) continuous review model for items with an exponential random lifetime and a general renewal demand process through a Markov process. We derive a fundamental rate conservation theorem and show that all the other system performance measures can be obtained easily through the expected reorder cycle length. This leads to a simple expression for the total expected long run cost rate function in terms of the expected reorder cycle length. Subsequently, we derive formulas for computing the expected cycle lengths for the general renewal demand as well as for a large class of demands characterized by the phase type interdemand time distribution. We show analytically when the cost as a function of the reorder level is monotone, concave, or convex. We also show analytically that, depending on the behavior of the expected reorder cycle, the cost as a function of the order‐up level is either monotone increasing or unimodal. These analytical properties enable us to understand the problem and make the subsequent numerical optimization much easier. Numerical studies confirm and illustrate some of the analytical properties. The results also demonstrate the impact of various parameters on the optimal policy and the cost. © 1999 John Wiley & Sons, Inc. Naval Research Logistics 46: 39–56, 1999     

2014nccet：一种新的用于减少远程Cache访问延迟的最后一次写访问预测方法*

为减少远程Cache访问延迟,提高共享存储系统的性能,提出了一种新的基于程序内在写突发特性的最后一次写访问预测方法,并对一个具体的目录协议进行了改造,以支持该预测方法。通过预测Cache块的最后一次写访问并提前对其进行降级,处理器能直接从主存中读取数据,从而减少了远程Cache访问所需的一个网络跳步数。与当前基于指令的预测方法相比,该方法能极大减少存储开销。基准测试程序的评测结果表明,该方法能获得83.1%的预测准确率,并且能提高8.57%的程序执行性能,同时与基于指令的预测方法相比,该方法能分别减少历史踪迹表69%的存储开销和签名表36%的存储开销。     

面向多核网络分组处理系统的线程亲和缓冲区管理机制

基于通用多核架构的网络分组处理系统性能受到诸如分组IO开销高、多核共享内存及进程调度竞争大、页表缓冲表项失效率高等问题的困扰。为此提出一种基于通用多核网络分组处理系统、面向高速分组转发应用的线程亲和缓冲区硬件管理机制,并在网络专用协处理引擎上实现。该机制采用无中断的线程亲和调度策略,将包含控制信息与缓冲区地址信息的描述符和分组数据按照分组处理的线程号链式地对应加载在多个地址连续的共享缓冲区中。基于通用多核和现场可编程门阵列平台进行报文转发测试,实验结果表明,采用线程亲和缓冲区管理机制能使平均报文转发处理性能提升12.4%,有效地降低IO开销和TLB表项失效率。     

A novel facial emotion recognition scheme based on graph mining

Recent years have seen an explosion in graph data from a variety of scientific, social and technological fields. From these fields, emotion recognition is an interesting research area because it finds many applications in real life such as in effective social robotics to increase the interactivity of the robot with human, driver safety during driving, pain monitoring during surgery etc. A novel facial emotion recognition based on graph mining has been proposed in this paper to make a paradigm shift in the way of representing the face region, where the face region is represented as a graph of nodes and edges and the gSpan frequent sub-graphs mining algorithm is used to find the frequent sub-structures in the graph database of each emotion. To reduce the number of generated sub-graphs, overlap ratio metric is utilized for this purpose. After encoding the final selected sub-graphs, binary classification is then applied to classify the emotion of the queried input facial image using six levels of classification. Binary cat swarm intelligence is applied within each level of classification to select proper sub-graphs that give the highest accuracy in that level. Different experiments have been conducted using Surrey Audio-Visual Expressed Emotion (SAVEE) database and the final system accuracy was 90.00%. The results show significant accuracy improvements (about 2%) by the proposed system in comparison to current published works in SAVEE database.