上海电网35kV变电站仿真系统及培训方案

本文介绍了一个实践中正在运行的35kV变电站——厦门站仿真培训系统。讨论了仿真系统的设计思想,仿真对象选择,仿真功能,硬件和软件系统方案,并对35kV变电站值班人员和无人值班变电站的运行操作人员培训方案进行设计。     

现实场景下行人身份识别

为了实现现实场景下的行人身份识别,结合人脸识别和行人重识别（Re-id）技术设计行人身份识别系统. 人脸识别技术能够在检测到行人的人脸时提高行人身份识别的精度,行人重识别通过改进的衣着信息消除（ICESD）模块来解决现实场景下的行人换装问题. 在存在大量未知行人的现实场景下,通过先识别是否为已知行人,然后识别具体行人类别的阈值过滤识别过程显著提升准度. 现实场景的应用测试和比较试验表明,相较于主流深度学习方法,在换装场景下,行人重识别技术能实现性能的显著提升,人脸识别也能显著提升整体识别准度. 系统的实时性分析表明,行人身份系统可以布置在现实场景下.     

创新用人机制促进企业发展

人才对企业的生存发展起着至关重要的作用，创新用人权制，企业需要在识人、选人、招人、用人等方面下功夫，并采取有效的激励措施。     

Algorithms for automated characterization of cell populations in thick specimens from 3-D confocal fluorescence microscopy data

Methods are presented for the automated, quantitative and three-dimensional (3-D) analysis of cell populations in thick, essentially intact tissue sections while maintaining intercell spatial relationships. This analysis replaces current manual methods which are tedious and subjective. The thick sample is imaged in three dimensions using a confocal scanning laser microscope. The stack of optical slices is processed by a 3-D segmentation algorithm that separates touching and overlapping structures using localization constraints. Adaptive data reduction is used to achieve computational efficiency. A hierarchical cluster analysis algorithm is used automatically to characterize the cell population by a variety of cell features. It allows automatic detection and characterization of patterns such as the 3-D spatial clustering of cells, and the relative distributions of cells of various sizes. It also permits the detection of structures that are much smaller, larger, brighter, darker, or differently shaped than the rest of the population. The overall method is demonstrated for a set of rat brain tissue sections that were labelled for tyrosine hydroxylase using fluorescein-conjugated antibodies. The automated system was verified by comparison with computer-assisted manual counts from the same image fields.     

Fuzzy genetic sharing for dynamic optimization

Recently,genetic algorithms(GAs) have been applied to multi-modal dynamic optimization(MDO).In this kind of optimization,an algorithm is required not only to find the multiple optimal solutions but also to locate a dynamically changing optimum.Our fuzzy genetic sharing(FGS) approach is based on a novel genetic algorithm with dynamic niche sharing(GADNS).FGS finds the optimal solutions,while maintaining the diversity of the population.For this,FGS uses several strategies.First,an unsupervised fuzzy clustering method is used to track multiple optima and perform GADNS.Second,a modified tournament selection is used to control selection pressure.Third,a novel mutation with an adaptive mutation rate is used to locate unexplored search areas.The effectiveness of FGS in dynamic environments is demonstrated using the generalized dynamic benchmark generator(GDBG).     

Adaboost和随机图划分的无监督图像分类

针对当前大多数无监督图像分类方法不能对每个图像类进行特征选择和自动确定图像类别的数量问题,提出一种基于Adaboost和随机图划分的无监督图像分类方法。该方法包括两个部分:1)将图像分类问题看做是一个自动的随机图划分问题,其中图的每一个顶点代表一幅图像,通过划分形成的子图代表了图像类。再采用Ada-boost算法对每一个形成的图像类进行特征选择,从而得到每类图像的表达模型。2)采用一种基于蒙特卡洛马尔可夫链(MCMC)的随机采样算法(SWC)来对图进行划分。相比传统的随机采样算法,SWC具有更快的收敛速度。在两个图像数据集上的实验结果表明,本文方法的分类性能明显优于其他现有的无监督分类法。     

无监督中文名词短语指代消解研究

为减小语料库对中文指代消解的影响,设计一个基于无监督聚类的中文名词短语指代消解平台并给出其预处理、特征选择及聚类过程。采用3种评测工具对中文新闻语料进行评测,在自动情况下,平均F值为59.43%。实验结果表明,该中文指代消解平台能够较好地解决中文缺少语料库的问题。     

基于无监督支持向量机的相似日选择

负荷预测是电力系统安全经济运行的基础,合理选择预测相似日有利于提高负荷预测的精度,而传统依据人工经验选择相似日并不具备最好的预测效果。为了进一步提高负荷预测准确度,该文深入研究了负荷预测中相似日选择的问题,依据对日负荷水平及趋势的影响大小提取特征量,并在此基础上提出一种基于无监督支持向量机的模型。最后,通过算例验证了所提模型的有效性。     

联合均等采样随机擦除和全局时间特征池化的视频行人重识别方法

针对为解决视频监控中遮挡、背景物干扰,以及行人外观、姿势相似性等因素导致的视频行人重识别准确率较低的问题,提出了联合均等采样随机擦除和全局时间特征池化的视频行人重识别方法。首先针对目标行人被干扰或部分遮挡的情况,采用了均等采样随机擦除（ESE）的数据增强方法来有效地缓解遮挡,提高模型的泛化能力,更准确地匹配行人;其次为了进一步提高视频行人重识别的精度,学习更有判别力的特征表示,使用三维卷积神经网络（3DCNN）提取时空特征,并在网络输出行人特征表示前加上全局时间特征池化层（GTFP）,这样既能获取上下文的空间信息,又能细化帧与帧之间的时序信息。通过在MARS、DukeMTMC-VideoReID 和PRID-2011三个公共视频数据集上的大量实验,证明所提出的联合均等采样随机擦除和全局时间特征池化的方法,相较于目前一些先进的视频行人重识别方法,具有一定的竞争力。     

Finding a person with a wearable acceleration sensor using a 3D position tracking system in daily environments

Person identification with accurate position information is essential for providing location-based services in real environments, such as a shopping mall. For this purpose, we propose a method that integrates 3D position information from environmental depth sensors and acceleration data from wearable devices to anonymously gather the trajectories of people who have wearable devices as well as others. Our proposed method identifies a person who has a wearable device by comparing two time series of acceleration data from device and position information. To do this, we extracted the behaviours of each axis using the changes of each bit of acceleration data at certain time periods. We evaluated our method with data collected at a shopping mall and a children’s playroom to investigate its effectiveness and robustness in different environments. Our evaluation results showed that it achieved an average identification of 85%, which is better than several alternative methods.