基于深度学习的ELM实时识别研究

基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。     

关于高师院校计算机专业教学改革的几点建议

本文以高等师范院校计算机专业教学改革为例,从教育观念的更新、教学模式的转变、教学手段的更新、教学对象的再认识、教学中的教材教法等多个方面分析以网络为主的现代教育模式改革的思想,目的在于探索出一条以多媒体和网络为主体的教学环境下的教育新思路。     

行波型热声发动机的分布参数法热力分析

在现有热声网络模型的基础上 ,对行波型热声发动机采用分布参数法进行了详细地分析 ,得到了较为完整的分布参数法网络模型 ,并给出了该模型的定性关系式 ,所得结果与采用集总参数法的结果有很大差异。分析表明 ,分布参数法更适合于一般的系统情况 ,集总参数法仅仅在一定条件下才适用。文中对一个实际系统应用分布参数法进行了分析 ,得到一些结论 ,对于我们设计行波型热声发动机有非常重要的理论指导意义     

Using activity focus networks to pressure terrorist organizations

The predominant idea for using network concepts to fight terrorists centers on disabling key parts of their communication networks. Although this counternetwork strategy is clearly a sound approach, it is vulnerable to missing, incomplete, or erroneous information about the network. This paper describes a different and complementary application of network concepts to terrorist organizations. It is based on activity focus networks (AFNs), which represent the complex activity system of an organization. An activity focus is a conceptual or physical entity around which joint activity is organized. Any organization has a number of these, which are in some cases compatible and in some cases incompatible. The set of foci and their relations of compatibility and incompatibility define the AFN. A hypothetical AFN for a terrorist organization is specified and tested in a simulation called AQAS. It shows that certain activity foci, and in particular one combination, have high potential as pressure points for the activity system. The AFN approach complements the counternetwork approach by reducing the downside risk of incomplete information about the communication network, and enhancing the effectiveness of counternetwork approaches over time. Steven R. Corman is Professor in the Hugh Downs School of Human Communication at Arizona State University and Chair of the Organizational Communication Division of the International Communication Association. His research interests include communication networks and activity systems, high-resolution text and discourse analysis, and modeling and simulation of human communication systems.     

神经网络在电磁场数值问题中的应用

针对目前电磁场数值处理中计算繁杂慢速和耗费资源过多的问题,本文提出了一种基于神经网络小波理论进行计算的新方法。文中着重介绍了利用小波变换神经网络的基本理论,结合电磁场数值计算的基本方法,推导变换进而得到一个通用公式和处理方法。在该方法中,神经网络学习内容少和数据处理简单,因此该方法具有计算速度快,占用资源少的优点,并有用硬件实现的前景。     

非网络版软件在Novell网上运行环境的设计

本文通过对非网络版软件在单机上运行环境分析，找出了一套调整此类软件使其在Ｎｏｖｅｌｌ网上运行的方法，并提出了非网络软件在Ｎｏｖｅｌｌ网上运行环境的设计的观点，具有较大的意义。     

Network externalities in a dynamic monopoly

We reconsider the role of network externalities in a dynamic spatial monopoly where the firm must invest in order to accumulate capacity, while consumers may have either linear or quadratic preferences. We (1) characterise saddle point equilibria, (2) prove that the extent of market coverage is increasing in the network effect and (3) unlike the existing static literature on the same problem, the monopolist may not make introductory price offers. Then, we briefly deal with the socially optimal solution, showing that, in general, a planner would serve more consumers than a profit-seeking monopolist.      

多时滞随机神经网络的稳定性

通过构建李雅普偌夫函数的方法和利用半鞅收敛定理对一类随机时滞神经网络的全局指数稳定进行了分析,提出了易于判定随机时滞神经网络几乎必然指数稳定性新的代数判据,推广了[1]中的主要结论.     

关于组合最优化理论的多媒体教学实践

本文讨论和分析了开发组合最比化理论系列课程多媒体课件的基本认识、课件的主要特色和体会，并提出了使用多媒体课件中应注意的问题．