Stepwise synthesis of cuprous oxide nanoparticles with adjustable structures and growth model

By stepwise adding of reducer N2H4·H2O,cuprous oxide(Cu2O)nanoparticles(NPs)with adjustable structures were synthesized.The features of Cu2O NPs were characterized by XRD,TEM and UV-Vis absorption spectra.When the reducer was added into the reactant system at one time,the sizes of the Cu2O NPs are in the range of 120–140 nm.Most Cu2O NPs are solid spheres.As the reducer was divided into two equal parts and stepwisely added,almost all the NPs are hollow spheres with good size(150–170 nm)distribution and dispersity.But when the reducer was divided into three or four equal parts and stepwisely added,the NPs are hollow spheres,core-shell structures or solid spheres,and the sizes distribution of the products is deteriorated.The effect of sodium hydrate(NaOH)was also probed.Addition of NaOH speeded up the nucleation and growth processes of Cu2O NPs.With the alkalinity increase,the shells of the hollow spheres become compact and the thicknesses of the shells increase,but the size distribution of the NPs is deteriorated.The absorption spectra of the Cu2O NPs are tunable.With the shell thicknesses increase,the absorption peaks have red shifts.An inside-outside growth model of Cu2O NPs was proposed to explain the results.The Cu2O single crystalline grains grow not only in the reactant solution,but also inside of the hollow nanospheres.The new Cu2O nanocrystallines can not only aggregate onto the shells of the nano hollow spheres,but also inside and outside of the hollow spheres,which leads to increasing the shell thicknesses of the hollow spheres,forming core-shell structures or small solid spheres of Cu2O NPs,respectively.     

基于SVM的老年痴呆症智能诊断研究

为了验证支持向量机（SVM）更适用于基于血常规数据的老年痴呆症的预测诊断,通过仿真实验,将BP神经网络、RBF神经网络、SVM支持向量机分别应用于老年痴呆症的预测诊断,建立3种算法对应的诊断模型,并对3种模型的预测结果进行分析比较,仿真实验在Matlab软件平台上进行. 结果表明,与BP、RBF神经网络方法相比,SVM模型预测准确度高,建模时间短,整体性能好,更适用于基于血常规数据的老年痴呆症预测诊断,实际应用时可以此结论作为理论指导.     

Parameter sensitivity and inversion analysis of a concrete faced rock-fill dam based on HS-BPNN algorithm

Considering the complex nonlinear relationship between the material parameters of a concrete faced rock-fill dam(CFRD) and its displacements, the harmony search(HS) algorithm is used to optimize the back propagation neural network(BPNN), and the HS-BPNN algorithm is formed and applied for the inversion analysis of the parameters of rock-fill materials. The sensitivity of the parameters in the Duncan and Chang's E-B model is analyzed using the orthogonal test design. The case study shows that the parameters φ0, K, Rf, and Kb are sensitive to the deformation of the rock-fill dam and the inversion analysis for these parameters is performed by the HS-BPNN algorithm. Compared with the traditional BPNN, the HS-BPNN algorithm exhibits the advantages of high convergence precision, fast convergence rate, and strong stability.     

基于改进的LSTM算法的时间序列流量预测

时间序列流量的预测问题是近年来机器学习的一个热点问题,通过改变长短期记忆网络(LSTM)层数、网络层神经元的个数、网络层之间的连接方式,特殊网络层的应用等网络结构以及优化器和损失函数的选择可以极大地提高预测的精度.本文提出多层LSTM算法,该算法是在传统LSTM算法上进行改进的单一模型,模型设计的复杂度低,可以提高机器学习的效率.模型采用一个输入层、5个隐藏层、1个输出层,同时包含1个全连接层和1个Dropout层,Dropout层的作用是防止机器学习过拟合.选择adam为模型优化器、mlse为模型损失函数、relu作为模型的激活函数.实验结果表明,与传统模型相比,该模型具有较好的泛化能力.     

An anisotropic micromechanical model for calculation of effective elastic moduli of Ni-based single crystal superalloys

An anisotropic micromechanical model based on Mori-Tanaka method is developed to calculate the effective elastic moduli of Ni-based single crystal superalloys. In the micromechanical model, the γ' precipitate with very high volume fraction is regarded as matrix, γ phase is divided into three parts as three different kinds of inclusions, and the actual cubic structure and orthogonal anisotropy properties of γ phase and γ′ precipitate are taken into account. Based on this anisotropic micromechanical model, the effective elastic moduli of Ni-based single crystal superalloys composite materials is obtained, and the influences of volume fraction and elastic constants of γ′ precipitate on the effective elastic moduli are also discussed. The results provide useful information for understanding mechanical behavior of composite materials in Ni-based single crystal superalloys and other anisotropic polygonal inclusion problem.     

基于3D-CBAM注意力机制的人体动作识别

针对已有的动作识别方法的特征提取不足、识别率较低等问题,结合双流网络、3D卷积神经网络和卷积LSTM网络的优势,提出一种融合模型. 该融合模型为了更好地提取人体动作特征,采用SSD目标检测方法将人体目标分割出作为局部特征和原视频的全局特征共同训练,并采用后期融合进行分类; 将3D卷积块注意模块采用shortcut结构的方式融合到3D卷积神经网络中,加强神经网络对视频的通道和空间特征提取; 并且通过将神经网络中部分3D卷积层替换为ConvLSTM层的方法,更好地得到视频的时序关系. 实验在公开的KTH数据集     

一种新的间断型备件需求预测方法

针对间断型需求因需求发生随机、需求量值波动大而导致预测困难这一问题,提出一种新的备件需求预测方法.该方法能分别预测需求发生时间和非零需求发生时的需求量值.对于0-1需求发生时间序列,采用调制方法对其进行平滑处理,运用神经网络对调制后的0-1时间序列进行预测,获得需求发生时间的预测值.采用时间聚合方法对实际备件需求时间序列进行预测,将滚动预测应用到解聚合过程中,得到备件的需求量预测值.使用三一重工砼活塞和核电设备的备件需求数据对方法进行验证,结果表明,该方法的预测精度要优于Croston方法、指数平滑法以及BP神经网络,证明了所提方法的有效性和准确性.     

Evaluate mobile video quality with LTE radio access network parameters

To evaluate the video quality, we tested sample videos delivered using HTTP adaptive streaming (HAS) in LTE network. In order to establish a correlation between radio access network (RAN) performance and quality of experience (QoE), we set up a testbed under different radio impairment conditions with three parameters: signal to interference and noise ratio (SINR), an amount of available network resource and a round trip latency. End users graded each video in a mobile equipment with their QoE Mearnwhile, we used a nonlinear model to simulate the comprehensive predicted mean opinion score (pMOS). Our results show that the nonlinear model can predict the enduser''s feedback. The pearson correlation coefficient (PCC) of the model is larger than 0.9. This demonstrate that the output of the model has a high correlation with the end users'' ratings and can reflect the QoE accurately. The method we developed will help mobile network operators evaluate the RAN performance of its QoE. It can also be used for HAS service to optimize LTE network and improve its QoE.     

Kriging模型在齿面磨损预测中的应用

为快速准确地对齿面磨损进行预测,考虑双齿啮合区的载荷分配并用Kriging方法建立了新的磨损数值仿真模型.基于Winkler弹性模型和轮齿啮合原理得到磨损量计算所需要的压力分布及啮合速度,在确定压力分布时考虑了由磨损带来间隙的影响,并对所需的载荷进行了动态分配;基于Archard磨损模型推导齿轮的磨损量数值仿真模型,得到了不同磨损次数下轮廓各个啮合点处的磨损深度;用Kriging方法和人工神经网络方法构建磨损与齿轮参数的关系代理模型,研究不同初始样本量下代理模型的逼近程度和拟合优度.算例计算结果表明:磨损量随磨损次数增加逐渐累积,参与啮合的齿廓各个位置的磨损量均不相同,节点处最小,越靠近齿根越大,主动轮(小齿轮)大于从动轮(大齿轮);综合比较3个初始样本量训练得到的Kriging模型表明,最小样本量为100时逼近程度和拟合优度都满足要求,并可预测未来的磨损量.采用Kriging模型具有较高的计算效率和精度,克服了磨损数值仿真模型计算耗时长的不足.     

液态挤压工艺ANN/GA建模与优化研究

利用人工神经网络方法(ANN)建立了工艺系统模型,用遗传算法(GA)对过程参数进行优化, 实验结果与预测值吻合良好, 为预测和控制该工艺成形质量提供了行之有效的手段。     

Communication simulation of on-board diagnosis network in high-speed Maglev trains

The on-board diagnosis network is the nervous system of high-speed Maglev trains, connecting all controller, sensors, and corresponding devices to realize the information acquisition and control. In order to study the on-board diagnosis network’s security and reliability, a simulation model for the on-board diagnosis network of high-speed Maglev trains with the optimal network engineering tool (OPNET) was built to analyze the network’s performance, such as response error and bit error rate on the network load, throughput, and node-state response. The simulation model was verified with an actual on-board diagnosis network structure. The results show that the model results obtained are in good agreement with actual system performance and can be used to achieve actual communication network optimization and control algorithms.

     

Electrical activities of neural systems exposed to sinusoidal induced electric field with random phase

The brain neural system is often disturbed by electromagnetic and noise environments, and research on dynamic response of its interaction has received extensive attention. This paper investigates electrical activity of Morris-Lecar neural systems exposed to sinusoidal induced electric field(IEF) with random phase generated by electromagnetic effect. By introducing a membrane depolarization model under the effect of random IEF, transition state of firing patterns, including mixed-mode oscillations(MMOs) with layered inter-spike intervals(ISI) and intermittency with a power law distribution in probability density function of ISI, is obtained in a single neuron. Considering the synergistic effects of frequency and noise, coherence resonance is performed by phase noise of IEF under certain parameter conditions. For the neural network without any internal coupling, we demonstrate that synchronous oscillations can be induced by IEF coupling, and suppression of synchronous spiking is achieved effectively by phase noise of IEF. Results of the study enrich the dynamical response to electromagnetic induction and provide insights into mechanisms of noise affecting information coding and transmission in neural systems.     

面向协作机器人的零力控制与碰撞检测方法研究

为了解决协作机器人柔顺交互控制问题，本文对机器人的零力控制和碰撞检测方法进行了深入研究。首先将逆运动学问题转化为（Newton-MP）广义逆和牛顿下山法的迭代求解问题。其次，针对协作机器人的零力控制问题，建立了基于速度三次摩擦力模型的完全动力学方程。对摩擦力模型进行遗传算法多参数辨识。再次，提出了基于One-Class卷积神经网络的碰撞检测方法，构建了无碰撞数据集，解决了传统碰撞检测方法建模不准确的问题。最后，通过实验证明，本文提出的Newton-MP优化方法具有良好的性能，绝对误差达到0.00013mm。与理想摩擦力模型进行对比，采用基于速度的三次摩擦力模型拟合出的摩擦力能够更好适用于零力控制。将外力矩观测器与One-Class卷积神经网络碰撞检测进行优缺点分析，可以证明One-Class卷积神经网络可以在不依靠模型的情况下，准确地检测机器人的异常碰撞。     

基于神经网络的VSLAM综述

传统的基于视觉的SLAM技术成果颇丰,但在具有挑战性的环境中难以取得想要的效果.深度学习推动了计算机视觉领域的快速发展,并在图像处理中展现出愈加突出的优势.将深度学习与基于视觉的SLAM结合是一个热门话题,诸多研究人员的努力使二者的广泛结合成为可能.本文从深度学习经典的神经网络入手,介绍了深度学习与传统基于视觉的SLAM算法的结合,概述了卷积神经网络(CNN)与循环神经网络(RNN)在深度估计、位姿估计、闭环检测等方面的成就,分析了神经网络在语义信息提取方面的优点,以期为未来自主移动机器人真正自主化提供帮助.最后,对未来VSLAM发展进行了展望.     

Crystal growth,structure and properties of bismuth triborate BiB<Subscript>3</Subscript>O<Subscript>6</Subscript> crystal

BiB₃O₆ (BIBO) single crystals with size of 46×23×10 mm³ and weight of 26.0 g have been successfully grown by top-seeded method. Problems encountered in the growth process of this crystal have been discussed in detail, and the methods of growing high-quality large crystals have been put forward. The relationship between their structure and properties is studied. The space group of monoclinic BiB₃O₆ is C2 and the cell parameters are a=7.1203(7) Å, b=4.9948(7) Å, c=6.5077(7) Å, β=105.586(8)″, and V=222.93(5) ų. The density of BIBO is 4.8965 g/cm³. The Mohs’s scale of hardness is 5.5–6. There is no cleavage face in the crystal. The transmittance of BIBO is about 80 percent in the range from visible coherent light to near-infrared light. The ultraviolet cutoff wavelength is at 276 nm. BiB₃O₆ is a biaxial crystal and has two sets of axes, and the relative orientation of (X, Y, Z) with regard to (a, b, c) is: X//b, (Y, c)=47.2°, (Z, a)=31.6°, determined by X-ray analysis combined with polarized microscopy. Second-harmonic-generation (SHG) experiments were carried out for the first time. In type I phase-matching (PM) directions (11.1°, 90°) and (168.9°, 90°), SHG conversion efficiencies of two directions for 1.064 μm light are up to 67.7% and 58%, respectively. We have also obtained the third-harmonic-generation (THG) of 1.064 μm. The comparative experiments between BIBO and KTP were carried out on conversion efficiency, transmittance and hardness. All the above results indicate that BiB₃O₆ is a kind of excellent nonlinear optical (NLO) crystal.     

Vacuum degree effects on betavoltaics irradiated by <Superscript>63</Superscript>Ni with differently apparent activity densities

For many current betavoltaics, beta sources and PN junction energy conversion units are separated. The air gap between the two parts could stop part of decay beta particles, which results in inefficient performance of the betavoltaic. By employing ⁶³Ni with an apparent emission activity density of 7.26×10⁷ and 1.81×10⁸ Bq cm^?2, betavoltaic performance levels were calculated at a vacuum degree range of 1×10⁵ to 1×10^?1 Pa and measured at 1.0×10⁵ and 1.0×10⁴ Pa, respectively. Results show that betavoltaic performance levels improve significantly as the vacuum degree increases. The maximum output power (P _max) exhibits the largest change, followed by short-circuit current (I _sc), open-circuit voltage (V _oc), and fill factor. The vacuum degree effects on I _sc, V _oc, and P _max of the betavoltaic with low apparent activity density ⁶³Ni are more significant than those of the betavoltaic with high apparent activity density ⁶³Ni. Moreover, the improved efficiencies of the measured performances are larger than the calculated efficiencies because of the low ratio of I _sc and reverse saturation current (I ₀). The values of I ₀, ideality factor, and shunt resistance were estimated to modify the equivalent circuit model. The calculation results based on this model are closer to the measurement results. The results of this research can provide a theoretical foundation and experimental reference for the study of vacuum degree effects on betavoltaics of the same kind.     

基于趋势检查法的遗传神经网络模型及工程应用

针对神经网络中模型可靠性问题,提出了趋势检查法的思路,采用评价指标中评价等级的影响趋势对模型进行检查,基本过程为不断调整模型参数、训练、趋势检查,直到获得最优模型。趋势检查法为一种通用方法,可用于任何基于先知经验方法的模型可靠性检查,为模型可靠性检查提供了一种新思路。对于神经网络学习样本贡献度不同的问题,采用样本加权的方法,对样本进行预处理,并将样本权值应用于神经网络的目标函数中,由此建立了加权神经网络目标函数。最后引入遗传算法来优化神经网络参数,建立了基于趋势检查法的遗传神经网络模型,并应用于实际工程中的围岩分类问题,结果表明该模型泛化能力强,具有较高的分类精度。     

基于PSENet的自然场景文字检测网络改进

由于自然场景中的图像背景复杂、文字排列不规则、光照条件不确定等因素文字检测难度较大,且传统检测方法的效果并不理想。在研究文字分割检测方法PSENet(Progressive Scale Expansion Network)的基础上,提出了一种针对自然场景文字检测的改进方法。该方法由卷积神经网络提取特征模块,再通过渐进式规模扩张对文字区域进行分割检测。改进点主要是使用高精度的语义分割网络RefineNet(Refinement Network)对卷积网络特征提取模块进行优化,且增加较多的残差连接及链式池化,提高网络对文字区域的检测精度。通过对数据集ICDAR2015的实验结果对比表明所提出的改进算法在精度方面略高于改进前,且能更好地解决文字粘连问题。     

基于遗传算法的BP神经网络在声音智能监控中应用

针对标准的BP神经网络对于声音信号在线监控模型的预测误差比较大,提出了一种用遗传算法优化BP神经网络的算法,建立了声音监控的预测模型。遗传算法优化BP神经网络主要是用遗传算法来优化BP神经网络的初始权值和阀值,然后通过训练BP神经网络以得到预测模型的最优解,优化后的神经网络具有预测误差比较小、反应速度快等特点。实验结果证明,利用遗传算法优化BP神经网络在声音的智能监控中取得了比较好的效果,达到了系统设计的目的。     

基于虚拟电价的电动汽车充放电优化策略

为了实现负荷侧的"削峰填谷"，在用户侧降低充电费用和电池损耗成本，提出了一种基于虚拟电价理论的电动汽车充放电优化策略。利用虚拟电价建立负荷侧的电价模型；利用动态分时电价建立用户侧的充电电价模型，将两个电价模型进行匹配整合，形成完整的电动汽车充放电的优化调度模型。为优化上述模型，确保整个调度的可实施性，在模型中使用BP神经网络和遗传算法进行预测与优化，同时使用小波分析和模糊聚类方法对充放电负荷进行去噪，并划分不同的电价时段，最后使用MATLAB软件对该模型进行了仿真验证。