可分离凸规划问题的交替邻近梯度法的次线性收敛率

给出了目标函数为3个凸函数的和且具有线性约束的可分离凸规划问题的交替邻近梯度法在遍历意义下的次线性收敛率为■的一个充分条件.     

一类Schrödinger-Poisson系统非平凡解的存在性

利用变分方法和临界点理论,研究了一类Schrödinger-Poisson系统,其中泊松项为更一般的形式,通过给非线性项加拟临界增长和AR条件,得到了该系统非平凡解的存在性。补充和推广了以往研究Schrödinger-Poisson系统的相关结果。     

The landing zone – Ground for model transfer in chemistry

This essay proposes a new notion - the landing zone - in order to identify conceptual features that allow modelers to transfer mathematical tools across disciplinary borders. This discussion refers to the transferable models as ‘templates’. Templates are functions, equations, or computational methods that are capable of being generalized from a particular subject matter. There are formal and conceptual prerequisites for the transfer of a template to a new domain. A landing zone is an ontology that contributes to the satisfaction of these conditions for successful transfer. This paper presents a case study on a model in chemistry - the Quantum Theory of Atoms in Molecules (QTAIM) - that makes use of transferred templates from physics - the virial theorem and the wave function. The landing zone in this case is a new ontological notion, that of the topological atom, which prepares ground for the use of the virial theorem and the wave function in chemistry. The virial theorem requires that there exists in-principle stability to the system that it represents, and the wave function requires transformation in its representation that is justified. The ontology of QTAIM - the landing zone for these templates - grounds the scientific use of these templates in the context of chemistry.     

数控机床故障分析与可靠性评价技术的研究

针对传统数控机床可靠性建模时只考虑故障间隔时间而忽略故障性质的现象，提出兼顾机械故障和电气故障的混合威布尔分布模型，进而提高模型准确度.利用最大似然法来确定混合威布尔分布模型参数，用皮尔逊相关系数来确定数控机床各故障时间之间的关系.并通过K-S方法对模型进行检验， 最后确定其符合混合威布尔分布，并运用可靠性评定方法得出数控机床的平均无故障时间.     

R型半胱氨酸体系电荷转移百分数的理论计算

采用密度泛函理论中的CAM-B3LYP方法, 在6-31G(d)基组水平上优化气相条件下R型半胱氨酸(R-Cys)分子的几何构型, 理论研究电子激发过程中R-Cys体系片段间的电荷转移特征, 并基于弛豫与非弛豫激发态密度计算片段间的电荷转移百分数. 结果表明: 对于SH片段, S0到S3的电荷转移百分数为57.96%； 对于COOH片段, S0到S1~S5各激发态的电荷转移百分数均为负值, 二者电荷转移的定性结果一致； 对于NH2片段, S0到S1和S4的电荷转移百分数分别为6.98%和31.45%.     

三目标优化:一种计算Pareto非劣解相对于各优化目标偏向度及其进一步分析的方法

考虑到理性决策者通常以获得高性价比结果为最佳选择,本文基于求解三目标优化问题得到的Pareto非劣解进一步分析.以"性价比"概念为基础,建立了Pareto前沿各点排序的基本规则,定义了相邻点概念,并明确了相邻点选择的方法.根据Pareto前沿各点与其相邻点的分布特点,计算得到了Pareto前沿各点的变化率;设计了灵敏比概念,得到了各Pareto非劣解相对各优化目标的偏向程度.本文的创新性贡献有3点:①利用三目标Pareto前沿灵敏比形成的新支配关系,进一步得到了比Pareto非劣解集范围更小的子集;②首次量化出三目标优化问题的Pareto非劣解相对于各优化目标的偏向度;③给出了各Pareto非劣解偏向于各优化目标的不平衡度,得到了不平衡度最小的解.最后,通过具体算例演示了上述计算过程,并与多种常用方法的计算结果进行了对比分析,验证了文中所述方法的可行性和有效性.本文研究成果对于进一步认识Pareto非劣解所具有的重要特性,深化三目标优化问题的求解是一次重要的理论推进.     

基于改进多目标遗传算法的连铸二冷过程优化

采用一种改进的多目标遗传算法对二冷工艺进行优化.改进的多目标遗传算法应用概率法选取选择算子，根据适应度值来动态计算交叉和变异概率，能够得到更好的全局最优解，提高算法精度和整体性能.在基于凝固传热模型的二冷优化过程中，采用变间距差分法离散求解传热方程，对比粒子群算法、多目标遗传算法，改进的多目标遗传算法搜索效率高，得到的价值函数最小.在实际生产中，采用优化后的二冷工艺，使得总用水量减少约10%，提高了铸坯质量，达到了节能降耗的要求.     

粉末冶金法制备混杂增强铜基复合材料的增强体竞争析出行为研究

Copper matrix composites reinforced by in situ-formed hybrid titanium boride whiskers (TiB_w) and titanium diboride particles (TiB_2p) were fabricated by powder metallurgy. Microstructural observations showed competitive precipitation behavior between TiB_w and TiB_2p, where the relative contents of the two reinforcements varied with sintering temperature. Based on thermodynamic and kinetic assessments, the precipitation mechanisms of the hybrid reinforcements were discussed, and the formation of both TiB_w and TiB_2p from the local melting zone was thermodynamically favored. The precipitation kinetics were mainly controlled by a solid-state diffusion of B atoms. By forming a compact compound layer, in situ reactions were divided into two stages, where Zener growth and Dybkov growth prevailed, respectively. Accordingly, the competitive precipitation behavior was attributed to the transition of the growth model during the reaction process.     

MADDPG算法并行优先经验回放机制

多智能体深度确定性策略梯度(multi-agent deep deterministic policy gradient, MADDPG)算法是深度强化学习方法在多智能体系统(multi-agent system, MAS)领域的重要运用,为提升算法性能,提出基于并行优先经验回放机制的MADDPG算法。分析算法框架及训练方法,针对算法集中式训练、分布式执行的特点,采用并行方法完成经验回放池数据采样,并在采样过程中引入优先回放机制,实现经验数据并行流动、数据处理模型并行工作、经验数据优先回放。分别在OpenAI多智能体对抗、合作两类典型环境中,从训练轮数、训练时间两个维度对改进算法进行了对比验证,结果表明,并行优先经验回放机制的引入使得算法性能提升明显。     

基于群偏好与用户偏好协同演化的群推荐方法

群推荐系统已经成为社交网络平台的重要工具,为群体用户提供兼顾个性化和整体满意度的产品和服务.现有群推荐方法大多是对个性化推荐方法的集成和聚合,忽略了群体和用户的交互影响以及群偏好和成员偏好的动态变化,从而无法保障群推荐系统的效果.为此,本文提出一种基于群偏好和用户偏好协同演化的群推荐方法,能够建模群体和用户的动态交互.具体而言,本文将用户偏好建模成其历史偏好和群影响的加权聚合结果,将群偏好建模成群历史偏好和新加入成员偏好的加权聚合结果,最终预测群体可能消费的产品列表和成员可能加入的群体列表.实验结果表明,本文所提模型在群体消费行为和用户加群行为的预测表现都优于基准算法,并兼具很好的鲁棒性.