卷积神经网络发展

卷积神经网络作为深度学习的重要代表算法之一,被应用到人脸识别、行为识别、字符识别和图像处理等多个领域,推动着人工智能的快速发展.本文在分析整理大量文献后,总结了卷积神经网络的发展阶段,介绍了卷积神经网络的基本结构和关键技术,选取LeNet、AlexNet、VGGNet、GoogLeNet和ResNet等几种具有代表性的网络进行分析对比,并指出各个网络的应用场景和优缺点,最后提出卷积神经网络未来的研究发展方向.     

模糊微分方程的反周期解

主要利用微分包含的方法得到模糊微分方程的反周期解的存在性结果，并给了一个例子说明主要结果的可行性.     

数据驱动的铝基复合材料性能预测和逆向设计

采用数据驱动的方法对 SiCp(0.5CNT)/7075Al 铝基复合材料的化学成分以及制备工艺进行了分析, 针对抗拉强度和延伸率两个力学性能进行了特征重要性分析, 构建了包含 8 种机器学习算法的集成框架, 自动进行模型的参数调优和最优模型选择, 并在此基础上进行了材料逆向设计. 实验结果表明, 在 470 ${^\circ}$C 固溶 40 min, 120${^\circ}$C 时效 15 h 的热处理工艺下, SiCp(0.5CNT)/7075Al-1.0Mg 复合材料抗拉强度和延伸率的预测值为 617.48 MPa 和 2.98%, 实验值为 647.0 MPa 和 3.31%, 两项物理性能的平均绝对百分比误差(mean absolute percentage errors, MAPE)较小, 依次为 4.56% 和 9.97%. 这说明本数据驱动方法对铝基复合材料的工艺优化和性能提升有一定指导意义.     

企业生命周期下投资者关注对权益资本成本的影响

基于行为金融学的价格压力假说，结合企业生命周期理论，以2007—2014年的主板上市公司为样本，分析投资者关注对权益资本成本的影响以及这种影响在企业生命周期中存在的阶段性差异。研究发现，投资者关注与权益资本成本显著负相关，即权益资本成本随着投资者关注的提高而降低。对于不同生命周期的企业，投资者关注对权益资本成本的影响存在差异。新生期、衰退期企业的权益资本成本并不会随着投资者关注程度的提高而降低；成长期、成熟期企业的权益资本成本随着投资者关注程度的提高而降低。可以看出，投资者能够识别企业所处的生命周期，并愿意为成长期、成熟期的企业支付更高的溢价。     

复掺配合比对自密实高流态混凝土性能影响

为了优化自密实高流态混凝土配合比,提高混凝土工作和力学性能,分别以胶凝材料、砂率和外加剂组分为关键变量,采用单一控制变量法,通过改变材料组分掺合量探究对自密实高流态混凝土性能影响.结果 表明:当胶凝材料在500～530kg/m3时,有较高流动性;超过530kg/m3时,间隙通过性效果好;低于530kg/m3时,可降低离析率.砂率为48％～50％时,拌合物流动性、间隙通过性较稳定,继续增大,抗离析性效果欠佳.当外加剂增大,流动性增加而抗离析性降低,外加剂为1.0％～1.2％时,对间隙通过性影响较小,与流动性、抗离析性较统一.随胶凝材料、砂率和外加剂各自增大,胶凝材料对3d、7d和28d抗压强度均有影响,砂率对3d抗压强度影响显著,外加剂对28d抗压强度影响明显.     

Superior corrosion resistance-dependent laser energy density in(CoCrFeNi)95Nb5 high entropy alloy coating fabricated by laser cladding

(CoCrFeNi)95Nb5 high entropy alloy(HEA)coatings were successfully fabricated on a substrate of Q235 steel by laser cladding technology.These(CoCrFeNi)95Nb5 HEA coatings possess excellent properties,particularly corrosion resistance,which is clearly superior to that of some typical bulk HEA and common engineering alloys.In order to obtain appropriate laser cladding preparation process parameters,the effects of laser energy density on the microstructure,microhardness,and corrosion resistance of(CoCrFeNi)95Nb5 HEA coating were closely studied.Results showed that as the laser energy density increases,precipitation of the Laves phase in(CoCrFeNi)95Nb5 HEA coating gradually decreases,and diffusion of the Fe element in the substrate intensifies,affecting the integrity of the(CoCrFeNi)95Nb5 HEA.This decreases the microhardness of(CoCrFeNi)95Nb5 HEA coatings.Moreover,the relative content of Cr2O3,Cr(OH)3,and Nb2O5 in the surface passive film of the coating decreases with increasing energy density,causing corrosion resistance to decrease.This study demonstrates the controllability of a high-performance HEA coating using laser cladding technology,which has significance for the laser cladding preparation of other CoCrFeNi-system HEA coatings.     

雷达导引头指向误差对导弹制导的影响与对策

受相控阵天线指向误差和天线罩瞄准误差的共同影响,相控阵雷达导引头存在较严重的指向误差。当指向误差斜率超出一定范围时会造成导弹制导系统出现寄生回路振荡问题,影响系统的稳定性和制导精度,在高空尤为明显。对此，本文通过构建相控阵雷达导引头制导系统模型,分析了导引头指向误差斜率对导弹制导的影响以及产生寄生回路振荡的机理;为消除导引头指向误差斜率对制导的不利影响,在指向误差测量补偿的基础上,又给出了一种多模型扩展卡尔曼滤波的估计方法,对导引头指向误差斜率进行实时估计与补偿。数字仿真结果表明，所提方法能够有效改善导引头指向误差斜率对制导系统的不利影响,提高系统稳定性和制导精度。     

A method for recovery of iron,titanium, and vanadium from vanadium-bearing titanomagnetite

An innovative method for recovering valuable elements from vanadium-bearing titanomagnetite is proposed. This method involves two procedures: low-temperature roasting of vanadium-bearing titanomagnetite and water leaching of roasting slag. During the roasting process, the reduction of iron oxides to metallic iron, the sodium oxidation of vanadium oxides to water-soluble sodium vanadate, and the smelting separation of metallic iron and slag were accomplished simultaneously. Optimal roasting conditions for iron/slag separation were achieved with a mixture thickness of 42.5 mm, a roasting temperature of 1200°C, a residence time of 2 h, a molar ratio of C/O of 1.7, and a sodium carbonate addition of 70 wt%, as well as with the use of anthracite as a reductant. Under the optimal conditions, 93.67% iron from the raw ore was recovered in the form of iron nugget with 95.44% iron grade. After a water leaching process, 85.61% of the vanadium from the roasting slag was leached, confirming the sodium oxidation of most of the vanadium oxides to water-soluble sodium vanadate during the roasting process. The total recoveries of iron, vanadium, and titanium were 93.67%, 72.68%, and 99.72%, respectively.     

基于变量选择-神经网络模型的复杂路网短时交通流预测

摘要： 针对传统交通流预测模型正在由单断面历史数据处理向多断面、多时刻历史数据处理转变,但在考虑各断面间的影响时,多变的交通状况往往会使预测模型复杂化的问题,引入一种多元线性回归最小绝对收缩和选择算子方法(Lasso),并利用其优秀的变量选择能力,在复杂路网多断面中选出相关性较高的断面；结合神经网络(NN)的非线性特性,提出了Lasso NN组合模型.结果表明：Lasso NN模型在路网交叉口对未来15 min交通流数据预测的误差率低于9.2%；在非交叉口的误差率低于6.7%,总体优于各自单独使用得出的结果.