糖尿病临床诊断事件序列中频繁模式的发现算法

随着各大医院信息化建设进程的不断推进,医院中的各生产系统如HIS（医院信息化系统）、EMR（电子病历系统）等已经积累了规模庞大的临床数据,这种临床大数据对于提升临床医疗质量有着深远的意义。糖尿病作为一种慢性病,容易引发多种并发症如肾病、眼病等。为了找出糖尿病并发症出现的规律,本文首先对糖尿病历史临床诊断数据进行事件序列化,然后对传统的SPADE算法进行改进,提出一种糖尿病临床诊断事件序列频繁模式发现算法NFPS,该算法考虑糖尿病治疗时间间隔,通过时间窗口的设定,支持对该时间窗口内糖尿病并发症频繁出现模式的发现。实验结果表明其在临床糖尿病并发症频繁模式发现上的有效性。     

基于OpenCL的人脸识别与隐藏系统

为加强人脸识别系统的隐私保护,本文引入基于灰度图像改变的加密解密算法对重要识别对象进行保护。为提高人脸识别与加密解密速度,运用OpenCL技术在异构平台下实现单设备与多设备的并行加速。测试结果表明,在使用单GPU时人脸识别局部加速达到1.9x,图像加密加速达到6.65x,而双GPU加速时,前者加速比可达4.11x,后者加速比与单GPU近似。      

耐高压大行程高速开关电磁铁的动态特性

由于传统以磁路计算为基础的动态特性计算精度较低,基于有限元方法建立了电磁铁的动态数学模型,并通过实验进行了验证,电流误差<4%,响应误差<6%,证明模型是可信的。在此基础上,对耐高压大行程高速开关电磁铁进行动态仿真,分析探讨了盆形极靴的结构参数对动态特性的影响规律。研究结果表明：为使电磁铁动态特性最优,盆口高、导套厚度、隔磁角和隔磁环长度有最佳值,径向间隙应尽可能小;考虑到耐压强度和成本,导套厚度和径向间隙均不能过小。结合实际设计要求,采用优化后的结构参数研制的耐高压大行程高速开关电磁铁在4 mm行程时的响应时间为2495 ms。     

应用神经网络决定多机系统快速汽门控制规律

提出了一种应用单隐层前向神经网络决定多机电力系统暂态稳定快速汽门 控制规律的方法。神经网络采用故障前各发电厂单台机组的有功功率和开机 台数等稳态量以及故障位置和类型等当地暂态信息作为输入，其输出为不控情 况下故障后某一适当时段内发电机转子间的最大相对摇摆角以及汽门关闭持续时 间的上、下限，前者用来与某一给定临界值进行比较以决定是否需要控制，后 者则用于确定合适的关闭持续时间。为了说明这种神经网络结构的合理性，文 中定性地证明了其输出与输入之间近似呈连续映射关系，并给出了一个6机2 2节点试例系统的结果。     

受扰轨迹的分群研究

以时域、频域和小波分析3种处理方法提取摇摆曲线的特征,对有限时段内稳定的多机摇摆曲线进行同调群识别,分离出所有潜在的失稳群。将摇摆曲线经过数字低通滤波器后重新采样,再求取各曲线的时域特征;用快速傅里叶变换得到其频域特征;用小波变换得到曲线在特定频段内的特征。然后用规则库分析这些特征,对摇摆曲线进行分群,其结果与EEAC算法对后继时段分析所得到的失稳模式基本吻合。     

Effect of Air Pressure on Hardened Layer of U75V 60 kg/m Heavy Rail after Heat Treatment

The samples cut from U75V 60 kg/m heavy rail are heated to 900 ℃ in resistance furnace and then put into air spraying channel to be cooled for 80 s, and change air pressure from 0.16 MPa to 0.33 MPa, and observe the effect of air pressure on hardened layer. The thickness and hardness of hardened layer increases with the increase of air pressure, and the thickness is more than 24 mm at the center and top fillets of rail head, and more than 15 mm at the blow fillets of rail head when air pressure is more than 0.26 MPa. During the tempering after heat treatment, tempering temperature of rail head is more than 570 ℃ when air pressure is separately 0.16 MPa, 0.20 MPa and 0.23 MPa, which is higher than finishing temperature of pearlite transformation at the cooling rate of 3 ℃/s according to CCT curve of U75V steel. When air pressure is separately 0.26, 0.30 and 0.33 MPa, the tempering temperature is 529 ℃ lower than finishing temperature of pearlite transformation at the cooling rate of 3 ℃/s. Under this condition, pearlite transformation is finished totally, so in order to reduce air consumption and control the cost, proper air pressure for U75V 60 kg/m heavy rail heat treating should be 0.26 MPa; the cooling rate increases with the increase of air pressure, and the average cooling rate on the surface of rail head is more than 3.21 ℃/s when air pressure is more than 0.26 MPa, and the largest cooling rate occurs at the top fillets of rail head.     

Using Walsh-m Composite Sequence and LDPC-QPSK Joint Iteration to Reduce ISI in Long-range Underwater Acoustic Systems

To reduce inter-symbol-interference (ISI) in underwater acoustic (UWA) communication systems, a method based on LDPC-QPSK joint iteration and Walsh-m composite sequence is proposed in this paper. The method is intended for use in long-range and low signal-to-noise ratio (SNR) UWA communications. At the transmitter, Walsh-m composite sequence is introduced to resist multipath effect. At the receiver, a soft-input soft-output (SISO) module is implemented in a joint iterative process between QPSK demodulator and LDPC decoder. This method is demonstrated in three types of UWA channel models: positive, negative and invariable sound velocity gradients channels. It is shown that through contrastive simulation experiments, this method is more efficient than conventional methods based on independent decoding and demodulation. After two rounds of joint iteration, the proposed method can obtain 2.5 dB over conventional method at BER of 10-5. Numerical results verify that the proposed method is a good candidate for long-range underwater acoustic communication systems.     

基于GPU平台的模乘算法实现

利用图形处理器硬件平台进行ECC椭圆曲线密码体制的有关计算及其相关攻击是一个较新的研究课题,其基础在于如何应用GPU硬件平台实现大整数模乘运算。文章针对NIST提出的素域F2192-264-1,基于GPU硬件的CUDA计算平台提出了实现该素域模乘算法的并行方法,详细说明了该方法在GPU上的数据组织结构和执行效率。实验数据表明,基于GPU硬件平台的模乘算法的速度约是用Mircal包计算该素域模乘的1 200倍,约是用GMP包的110倍。     

Application of fractal method to optimum and decoupling design of electromechanical system

ＡＰＰＬＩＣＡＴＩＯＮＯＦＦＲＡＣＴＡＬＭＥＴＨＯＤＴＯＯＰＴＩＭＵＭＡＮＤＤＥＣＯＵＰＬＩＮＧＤＥＳＩＧＮＯＦＥＬＥＣＴＲＯＭＥＣＨＡＮＩＣＡＬＳＹＳＴＥＭＺｈｏｎｇＪｕｅＹｉＹｏｕｐｉｎｇＣｈｅｎＪｉｅ（ＣｏｌｅｇｅｏｆＭｅｃｈａｎｉｃａｌａｎｄ...     

桥梁振动特性的计算和探讨

在采用有限元对桥梁进行结构分析时，准确地确定桥梁的振动特性是非常重要的，本文简要地介绍了计算桥梁振动特性的方法，并详细介绍了如何简便地考虑桩土的作用，通过对计算的结果与实测的结果比较，证实了本文给出的计算方法的准确性可以满足工程的要求。