喷动床反应模型的研究进展

喷动床是一种可实现热量重复循环的高效反应器 .介绍了目前喷动床研究中出现的各种反应模型 ,从等温和非等温两个方面讨论了一维轴向、流管、半分割和流动反应耦合等各类模型的基本思想和理论及各自的特点 ,并提出了进一步发展的方向 .     

石蜡沉积对储层渗透率的影响研究

油气开采过程中的蜡质、沥青质等有机固相沉积。严重影响着油田的正常生产。固相沉积物不仅堵塞油井、输油管线,而且也给油藏多孔介质造成一系列的伤害。如堵塞油藏孔隙、降低孔隙度和渗透率,改变油藏岩石的润湿性,从而影响原油采收率的提高。本文采用热力学理论建立了液固两相相平衡方程,模拟计算石蜡沉积量对储层孔隙度和渗透率影响,进而求出对油井产能的影响。     

视觉特征对比解耦的广义零样本学习

广义零样本学习通常利用在ImageNet上预训练的深度模型来提取相应的视觉特征,然而预训练模型提取到的视觉特征不可避免地包含和语义无关的信息,这将导致语义—视觉对齐的偏差以及对不可见类的负迁移,从而影响分类结果。为解决上述问题,提出了视觉特征对比解耦的广义零样本学习模型(visual feature contrast decoupling for generalized zero-shot learning, VFCD-GZSL),通过解耦出视觉特征中的语义相关表示来降低冗余信息对分类结果的影响。具体来说,首先用条件变分自编码器生成不可见类的视觉特征。然后通过解耦模块将视觉特征解耦语义相关和语义无关的潜层表示,同时添加总相关惩罚和对比损失来鼓励两者间的相互独立,并用语义关系匹配模型衡量其语义一致性,从而指导模型学习语义相关表示。最后使用特征细化模块细化后的特征和语义相关表示联合学习一个广义零样本学习分类器。在四个数据集上的实验均取得较优的结果,证实了所提方法的有效性。     

图表示学习方法研究综述

针对图表示方法的相关解析任务进行了研究,从形式化定义出发,首先以不同核心技术作为分类标准将图表示学习方法划分为五大类,其包括基于降维解析、矩阵分解、随机游走、深度学习和其他表示学习方法。其次通过归纳与对比分析梳理各类技术发展脉络,进而深层次展现各类图表示方法的优劣。随后结合图表示学习的常用数据集、评估方法和应用领域的归纳分析,展开动态性、可扩展性、可解释性和可解析性的四维剖析。最后总结并展望了图表示学习的未来研究趋势与发展方向。     

硬约束限制的情感文本生成方法研究

预训练语言模型在情感文本的生成任务中取得了良好效果,但现有情感文本生成方法多使用软约束的方式控制文本整体的情感属性,缺乏单词和短语级别的硬性控制。为解决以上问题,提出硬约束限制下的情感文本生成方法。首先使用方面情感分析技术提取句子的方面词、情感词并判断情感极性;之后,选择目标情感的方面词和情感词作为预训练语言模型的硬约束输入来重建完整句子,其中,设计了一种新的单词权重计算方法,旨在使模型优先生成重要单词。实验结果表明,该方法生成的句子不仅具有方面级情感,在文本质量和多样性的评价指标上也有显著提高。     

TGA analysis of γ‐irradiated linear low‐density polyethylene

Linear low‐density polyethylene (LLDPE), based on butene‐1 or hexene‐1, was irradiated with γ‐rays under vacuum or in the presence of air. The study focused on the influence of the dose rate and the γ‐dose on the thermal properties of LLDPE. Differential scanning calorimetry, thermogravimetric analysis (TGA), and TGA/FTIR techniques were used to address the thermal behavior as a result of γ‐irradiation. During this irradiation, competition between crosslinking and scission reactions, subsequent to oxidation reactions, occurred in the polymeric material, which strongly depends on the experimental conditions. A decrease of the crystallinity for γ‐irradiated samples was observed in particular for samples irradiated under vacuum. This observation may be explained by increased hindrance of segment mobility due to crosslinking reactions that prevent crystal growth. TGA investigations revealed an enhancement of the thermal stability for samples irradiated under vacuum but not for those irradiated in air. © 2006 Wiley Periodicals, Inc. J Appl Polym Sci 100: 2790–2795, 2006     

Presumed joint probability density function model for turbulent combustion

A presumed joint probability density function (pdf) model of turbulent combustion is proposed in this paper. The turbulent fluctuations of reactant concentrations and temperature are described using a presumed joint pdf of three-dimensional Gaussian distribution based on first and second-order moments of reactant concentration and temperature. Mean reaction rates in both premixed and diffusion combustion are obtained by mean of integration under the presumed joint pdf. This model is applied to predict turbulent premixed combustion of sudden-expansion flow and turbulent jet diffusion methane/air flame. For turbulent premixed combustion, the predicted results of temperature distribution and maximum temperature using the proposed model agree better with the experiment than that using the conventional eddy-breakup (EBU)-Arrhenius model. For the turbulent jet diffusion methane/air flame, the predicted results of velocity, temperature and species concentrations using the proposed model, the Arrhenius, EBU-Arrhenius, and laminar flamelet models are compared with experiment data. Results obtained with the presumed pdf model and that obtained by the laminar flamelet model both agree well with experiments, while results using the other models have a significant difference. The presumed joint pdf model is used to predict the NO formation process, which also agrees well with the experiment data. A unified turbulent combustion model, in which both effects of turbulent diffusion and chemical dynamics are considered, is established for both premixed and diffusion combustion, especially for the process of NO formation.     

基于引力搜索算法的分数阶变异时序回归GSA-TSGM(1,1)模型

为了利用分数阶累加算子在灰色短期预测中的高效性能,首次将分数阶累加算子引入变异时序回归模型以期取得更高的预测精度。主要方法如下：首先取湖北省链子崖某监测点1978—1987年的十年数据作为训练集并使用引力搜索算法确定最佳分数阶累加阶数,而1988—1993年的六年数据作为验证集验证提出的模型;其次对比了经典灰色模型GM(1,1)、分数阶累加灰色模型、变异时序回归模型TSGM(1,1)三种灰色模型。结果如下：首先修正了陈西江等人变异时序回归模型仿真时出现的错误,其次表明了相比于其他的模型,基于引力搜索算法的分数阶累加时序回归模型在进行灰色长期预测中具有较高的预测精度。因此,通过分数阶累加算子提高了灰色理论中长期预测模型的精度,为灰色长期预测提供了指导。     

一种结合主题模型的推荐算法

针对传统协同过滤推荐算法存在的冷启动、数据稀疏以及相似度度量的准确性问题,基于LDA主题模型对文本隐式主题挖掘的有效性和KL散度在主题分布相似性度量的准确性,提出了结合LDA主题模型的矩阵分解推荐算法。首先,利用改进的LDA算法输出项目-主题分布,并用困惑度作为主题数设置的修正函数;然后分别基于余弦相似度和KL散度计算得到项目相似度矩阵,将得到的相似度矩阵结合原评分训练集输出预评分,再将预评分填充到训练集;最后将训练集输入ALS矩阵分解算法得到推荐结果。通过MovieLens数据集的实验结果表明,该算法在不同隐式参数设定下均能得到比ALS推荐算法以及更小的预测误差,并且最优预测误差小于传统推荐算法。该实验说明了通过集成LDA主题模型的ALS算法效果要优于其他推荐算法。     

Simulation of swirling combustion and NO formation using a USM turbulence-chemistry model

A unified second-order moment (USM) turbulence-chemistry model is used to simulate methane-air swirling combustion and NO formation for different swirl numbers. The simulation results are compared with those using the EBU-Arrhenius (E-A) combustion model, the simplified PDF model of NO formation in turbulent flows and the corresponding experimental results. The comparison indicates that the USM model is obviously better than the E-A model and the simplified PDF model. The E-A model cannot reasonably simulate the finite-rate kinetics, while the simplified PDF model, using a product of two single-variable PDF's instead of a joint PDF, remarkably under-predicts the NO reaction rate. The USM model gives the best agreement with the experimental results. Predictions show that as the swirl number increases the total NO formation at first decreases and then increases, which is in agreement with the experimental results.