基于SOM聚类和ECA的超短期光伏预测组合模型研究

为提高不同天气类型下光伏输出功率的预测精度,提出了一种基于注意力机制的超短期光伏预测组合模型.首先,通过皮尔逊相关系数分析,选取与光伏发电功率密切相关的关键气象因子,并对其进行逐月标准化,然后加权求和计算得到分类指标天气条件因子(Sky Condition Factor,SCF),以降低输入变量的维度,并消除季节对天气分类的干扰和众多气象因子之间的耦合关系.其次,通过自组织映射神经网络(Self-Organizing Map,SOM)对SCF进行无监督聚类,划分出3种天气类型.然后,在3种天气类型下分别构建卷积神经网络(Convolutional Neural Network,CNN)预测模型,并引入高效通道注意力模块(Efficient Channel Attention,ECA),自适应地为特征信息的多重通道分配相应的权重,使模型集中于重要的特征信息,提高模型的预测精度.采用历史实测数据进行仿真,结果表明:与未引入ECA模块的预测模型相比,所提预测模型在3种天气类型下的预测准确度分别提高了 1.006 1％,1.626 1％和1.610 4％,验证了该模型的有效性.     

基于HRCenterNet模型改进的朝鲜语古籍文字检测方法

为减少朝鲜语古籍中的小文字错检和漏检问题,提出了一种基于HRCenterNet模型改进的朝鲜语古籍文字检测方法.首先,将HRCenterNet中Bottleneck模块的3×3卷积运算替换为Involution算子,即将Bottleneck模块替换为Involution - Bottleneck模块.其次,通过引入ECA(efficient channel attention)注意力机制扩展Involution - Bottleneck模块,并由此提出了基于IENeck模块的HRCenterNet改进模型.最后,利用朝鲜语古籍数据集对改进的HRCenterNet模型和原模型分别进行了训练,并测试了其在不同IOU下的准确率、召回率以及F1等指标.实验结果表明,在IOU ≥ 0.6时,改进的HRCenterNet模型在朝鲜语古籍数据集上的准确率、召回率和F1指标均优于原模型,且IOU值越高模型的检测效果越好.这表明改进的HRCenterNet模型显著优于原模型,可应用于朝鲜语古籍文字的检测中.     

覆盖L 波段的宽带隐身雷达天线罩设计

设计了一款极化和角度不敏感的宽带频选吸波体,在X 波段实现了宽带透射窗口以及包含L波段在内的宽频带吸收。该频选吸波体采用频率选择表面与电磁超材料吸波体相结合的方式,通过级联加载多层耦合型频率选择表面和双层高阻表面完成总体结构设计。理论上,利用等效电路法对多层耦合型频率选择表面及整体结构展开分析,论证频选结构和频选吸波体的谐振机理。数值仿真结果显示,该频选吸波体可以实现8.1～11.7GHz频段内的宽频带透射,以及1.18～4 GHz、15～18 GHz 频段内的宽频带吸收,其中透射窗口的插入损耗不大于3 dB。将该频选吸波体作为平面雷达天线罩与微带天线相结合,分析天线的辐射性能和散射特性。研究结果显示,天线在工作频带内保持了良好的辐射性能,而在带外实现了RCS 的有效缩减,达到了天线系统的隐身目的。     

具有自校正与注意力机制相结合的场景文本检测北大核心CSCD

在日常生活中,存在着丰富的文本信息,对这些信息的提取,能够极大地提高人们的生活品质。但自然场景中文本信息表达形式丰富多样,文本形状各异,在检测过程中存在误检、文本区域定位不准问题。针对以上不足,本文提出了一种具有自校正与注意力机制相结合的文本检测方法。首先,在ResNet50骨干网络中嵌入自校正卷积(self-calibrated convolution, SConv)及高效通道注意力(efficient channel attention, ECA),使网络能够校正全局无关信息的干扰,并集中关注于文本区域,提取更加丰富的语义信息;其次,在特征融合后加入协调注意力(coordinate attention, CA),纠正不同尺度的特征图在融合过程中产生的位置偏差。最后,通过修正后的特征图预测得到多个不同尺度的文本实例,采用渐进尺度扩展算法,求出最终检测到的文本实例。实验结果表明,在任意方向数据集ICDAR2015以及弯曲文本数据集Total-Text、SCUT-CTW1500上,相比于改进前的ResNet50综合指标F值分别提升了1.0%、5.2%、5.4%,证明了本方法具有良好的检测能力。     

基于ECA规则的适应性工作流技术研究

工作流管理系统能够给企业的业务运行提供有力支持。但是,目前的工作流管理系统对过程模型的表达过分简化,影响了它在实际应用中的效果。为了提高工作流管理系统的适应性,文中形式化地定义了适合于工作流的ECA规则。在此基础上提出了基于ECA规则的工作流模型,并将该模型与基于Petri网和基于UML活动图的工作流模型进行了比较。提出了一种基于事件表达式重写策略的事件检测方法,以提高工作流引擎的执行效率。最后,介绍了一个基于ECA规则的工作流技术的应用系统。     

柔性工作流动态行为建模方法

针对现有工作流动态行为建模柔性差的问题,研究工作流动态行为的柔性定义方法。将活动的动态行为分为自身行为、嵌套行为和协作行为。自身行为实际上是活动的状态转移的过程,通过ECA规则描述活动的状态转移图可实现自身行为的定义;活动间协作行为实质上是参与活动间状态变化关联的过程,通过定义活动间状态的关联关系并用ECA规则描述协作语义,可实现对嵌套和协作行为的细粒度柔性定义。文中提出的可扩展方法适用于复杂或特殊环境下的流程建模。     

基于XML和ECA规则的工作流建模及执行

长期以来,工作流产品互操作性的缺乏、尤其是工作流建模语言互操作性缺乏,已成为阻碍工作流技术进一步发展的重要因素,XML作为万维网组织规定的结构化的标记语言,它可以用作描述过程模型的共同语法格式,有望解决工作流描述语言的互操作性问题。文章介绍了采用基于XML的工作流描述语言XFL及其工作流管理系统XFMS。由于XFL本身并不是可执行模型,XFMS系统在低层采用了基于规则的工作流执行系统。并给出了从XFL语言到ECA规则的映射方式,所采用的方法综合了面向规则系统的灵活性以及基于XML模型的可交换性和易学性。     

电子封装无铅化技术进展(待续)

介绍了两种用于电子封装行业的无铅技术—无铅焊料和导电胶连接技术,以及这两种技术的比较和应用,指出了无铅技术的若干前沿问题。     

Priority assignment in real-time active databases

Active databases and real-time databases have been important areas of research in the recent past. It has been recognized that many benefits can be gained by integrating real-time and active database technologies. However, not much work has been done in the area of transaction processing in real-time active databases. This paper deals with an important aspect of transaction processing in real-time active databases, namely the problem of assigning priorities to transactions. In these systems, time-constrained transactions trigger other transactions during their execution. We present three policies for assigning priorities to parent, immediate and deferred transactions executing on a multiprocessor system and then evaluate the policies through simulation. The policies use different amounts of semantic information about transactions to assign the priorities. The simulator has been validated against the results of earlier published studies. We conducted experiments in three settings: a task setting, a main memory database setting and a disk-resident database setting. Our results demonstrate that dynamically changing the priorities of transactions, depending on their behavior (triggering rules), yields a substantial improvement in the number of triggering transactions that meet their deadline in all three settings. Edited by Henry F. Korth and Amith Sheth. Received November 1994 / Accepted March 20, 1995     

Complex RFID event processing

Advances of sensor and radio frequency identification (RFID) technology provide significant new power for humans to sense, understand and manage the world. RFID provides fast data collection with precise identification of objects with unique IDs without line of sight, thus it can be used for identifying, locating, tracking and monitoring physical objects. Despite these benefits, RFID poses many challenges for data processing and management: (i) RFID observations have implicit meanings, which have to be transformed and aggregated into semantic data represented in their data models; and (ii) RFID data are temporal, streaming, and in high volume, and have to be processed on the fly. Thus, a general RFID data processing framework is needed to automate the transformation of physical RFID observations into the virtual counterparts in the virtual world linked to business applications. In this paper, we take an event-oriented approach to process RFID data, by devising RFID application logic into complex events. We then formalize the specification and semantics of RFID events and rules. We discover that RFID events are highly temporal constrained, and include non-spontaneous events, and develop an RFID event detection engine that can effectively process complex RFID events. The declarative event-based approach greatly simplifies the work of RFID data processing, and can significantly reduce the cost of RFID data integration. This work was done by F. Wang while working at Siemens Corporate Research. This work was done by S. Liu while visiting Siemens Corporate Research.