二冷配水优化建模与混合自适应PSO算法求解

针对连铸二冷区生产环境复杂且存在着大量水雾干扰的情况,建立了连铸水量优化模型并提出了一种混合的自适应粒子群算法来求解连铸二冷水优化问题。依据冶金过程中的工艺要求建立了二冷水量优化模型,并在经典的PSO算法基础上提出了适合该问题求解了混合自适应PSO算法。由于连铸过程存在着偏微分方程约束,传统的优化方法容易陷入局部最优解,不能达到很好的动态优化效果。研究了粒子群算法,基于种群的多样性,不断的自适应的更新粒子群算法中参数,将禁忌搜索的方法和传统的粒子群算法结合,增强了算法的局部搜索能力和全局寻找全局最优的能力。将该算法应用到连铸二冷水动态优化中,实验结果表面该算法能够快速有效的求解该优化问题。该方法用于连铸二冷水优化是可行的、有效的。     

有机化学实验在线开放课程设计的探索与实践

针对以往有机化学实验教学中存在的问题及学生所反映的改进期望,在有机化学实验在线开放课程的开发中,充分利用在线课程与网络教育平台,制作了针对性的教学微视频课程和相应的课件,建立了一套完整的测试模式与题库。通过考测机制,促进和强化学生的实验课前预习。而学生在充分预习的基础上,实现实验预习–现象记录–实验报告完整规范化和一体化,相应提高了学习效率。同时,建立了一套相对客观的成绩评估体系并取得了积极的成果。     

基于实例的检索

基于实例的推理是人工智能中一项重要推理技术,而推理的核心是检索.介绍了产品设计的相似度计算方法,以及实例特征参数权系数值的确定方法,最后给出产品设计实例库三级检索算法.     

基于麻雀搜索算法结合深度前馈神经网络的近红外模型转移方法研究

该文提出了一种基于麻雀搜索算法结合深度前馈神经网络（SSA-DFN）的近红外光谱模型转移方法。使用深度前馈神经网络拟合不同仪器采集到的光谱之间的非线性函数映射,并将麻雀搜索算法用于网络各层连接权值和阈值的初始化,通过种群中个体位置的迭代更新,求得连接权值和阈值的最优初始值;通过多次调整深度前馈神经网络模型的超参数,使网络拟合效果趋于最优,最终确定转移函数。为验证方法的有效性,分别从烟叶近红外光谱谱图、主成分投影和预测结果的角度,将SSA-DFN方法与分段直接校正算法（PDS）、典型相关性分析算法（CCA）转移前后的效果进行了对比。结果表明SSA-DFN方法转移后的从机光谱与原主机光谱重合度最高,转移后主、从机总糖、烟碱含量的预测结果差异不显著,预测平均误差从8.32%、9.15%分别降至4.65%、4.82%,预测均方根误差（RMSEP）和决定系数（R2）等指标均优于PDS和CCA,取得了最佳的转移效果,可满足企业需求。结果表明该方法是一种有效的模型转移方法。     

An Efficient SCF-PEKS Without Random Oracle Under Simple Assumption

The Public key encryption scheme with keyword search (PEKS), firstly put forward by Boneh et al., can achieve the keyword searching without revealing any information of the initial data. However, the original PEKS scheme was required to construct a secure channel, which was usually expensive. Aimed at resolving this problem, Baek et al. put forward an improved scheme, which tried to construct a Secure channel free PEKS (SCF-PEKS). Subsequently, several SCF-PEKS schemes were proposed, however most of them turned out only secure in the random oracle model, which possibly lead to the construction of insecure schemes. Therefore, Fang et al. put forward an enhanced SCF-PEKS construction, which was provably secure in the standard model, however this construction needed a strong and complicated assumption. Then Yang et al. put forward an SCF-PEKS construction under simple assumption, but their construction had a big reduction in efficiency. In this article, we propose an SCF-PEKS construction, which is provably secure under the same assumption as that of Yang et al.'s scheme, however, with better performance. Then we give its full security proof, along with the performance analysis. Finally, we improve the SCF-PEKS construction to resist Keyword guessing attack (KGA) and give its security demonstration.     

基于快速神经网络架构搜索的鲁棒图像水印网络算法

为解决深度学习在图像水印算法中计算量大且模型冗余的问题,提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性,提出基于快速神经网络架构搜索（neural architecture search,NAS）的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法,在预设的搜索空间中搜索最优网络结构,进行图像水印的高效嵌入与鲁棒提取。首先,将子网络中线性连接的全卷积层设置为独立的神经单元结构,并参数化表示结构单元内节点的连接,预先设定结构单元内每个神经元操作的搜索空间;其次,在完成一个批次的数据集训练后,依据神经元操作中的被采样次数和平均损失函数值动态更新概率;最后,重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上,大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑,本文算法具有较高的时间性能和较好的实验效果,在隐藏图像时,对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验,对比发现,本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行（列）等攻击优势显著;在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著,特别是对随机移除行（列）和椒盐噪声有较强的鲁棒性。     

有机环境污染物紫外光谱检索的神经网络方法

详细讨论了网络优化参数、模拟的测量过程中噪声及杂质对网络收敛性能及预测误差的影响。为加速网络收敛,提高紫外光谱检索的正确率,采用了导数光谱对反向传播的人工神经网络(BP-ANN)进行训练和检索,该方法对检索光谱中噪声、杂质,尤其是斜坡背景的允许程度明显提高。文章还将ANN方法与普通的相关系数法的识别结果进行了比较。结果表明,优化参数下的人工神经网络的库检索法在抗噪、容杂等方面都明显地优于普通的相关系数法,是一种很有效的紫外库检索方法。     

NANOMECHANICAL MAPPING OF CARBON BLACK REINFORCED NATURAL RUBBER BY ATOMIC FORCE MICROSCOPY

Atomic force microscopy （AFM） has the advantage of obtaining mechanical properties as well as topographic information at the same time. By analyzing force-distance curves measured over two-dimensional area using Hertzian contact mechanics, Young＇s modulus mapping was obtained with nanometer-scale resolution. Furthermore, the sample deformation by the force exerted was also estimated from the force-distance curve analyses. We could thus reconstruct a real topographic image by incorporating apparent topographic image with deformation image. We applied this method to carbon black reinforced natural rubber to obtain Young＇s modulus distribution image together with reconstructed real topographic image. Then we were able to recognize three regions; rubber matrix, carbon black （or bound rubber） and intermediate regions. Though the existence of these regions had been investigated by pulsed nuclear magnetic resonance, this paper would be the first to report on the quantitative evaluation of the interfacial region in real space.     

Raman Optical Activity of N-Acetyl-L-Cysteine in Water and in Methanol: The “Clusters-in-a-Liquid” Model and ab Initio Molecular Dynamics Simulations

Raman and Raman Optical Activity (ROA) spectra of N-acetyl-L-cysteine (NALC), a flexible chiral molecule, were measured in water and in methanol to evaluate the solvent effects. Two different solvation approaches, that is, the DFT based “clusters-in-a-liquid” solvent model and the ab initio molecular dynamics (AIMD) simulations, were applied to simulate the Raman and ROA spectra. Systematic conformational searches were carried out using a recently developed conformational searching tool, CREST, with the inclusion of polarizable continuum model of water and of methanol. The CREST candidates of NALC and the NALC-solvent complexes were re-optimized and their Raman and ROA simulations were done at the B3LYP−D3BJ/def2-TZVP and the B3LYP-aug-cc-pVDZ//cc-pVTZ levels. Also, AIMD simulations, which includes some anharmonic effects and all intermolecular interactions in solution, were performed. By empirically weighting the computed Raman and ROA spectra of each conformer, good agreements with the experimental data were achieved with both approaches, while AIMD offered some improvements in the carbonyl and in the low wavenumber regions over the static DFT approach. The pros and cons of these two different approaches for accounting the solvent effects on Raman and ROA of this flexible chiral system will also be discussed.     

Identification of methylated regions with peak search based on Poisson model from massively parallel methylated DNA immunoprecipitation-sequencing data

DNA methylation is one of the most important epigenetic modification types, which plays a critical role in gene expression. High efficient surveying of whole genome DNA methylation has been aims of many researchers for long. Recently, the rapidly developed massively parallel DNA‐sequencing technologies open the floodgates to vast volumes of sequence data, enabling a paradigm shift in profiling the whole genome methylation. Here, we describe a strategy, combining methylated DNA immunoprecipitation sequencing with peak search to identify methylated regions on a whole‐genome scale. Massively parallel methylated DNA immunoprecipitation sequencing combined with methylation DNA immunoprecipitation was adopted to obtain methylated DNA sequence data from human leukemia cell line K562, and the methylated regions were identified by peak search based on Poison model. From our result, 140 958 non‐overlapping methylated regions have been identified in the whole genome. Also, the credibility of result has been proved by its strong correlation with bisulfite‐sequencing data (Pearson R²=0.92). It suggests that this method provides a reliable and high‐throughput strategy for whole genome methylation identification.