面向负载均衡的高铁路网列车开行方案优化方法

针对当前高速铁路运营过程中存在的运输需求与运力资源不匹配现象,面向负载均衡原理研究了路网条件下运能可适配的高速铁路旅客列车开行方案优化与评估方法.首先,针对路网条件下列车开行方案优化,构建以提升经济效益、社会效益和网络负载均衡为目标的非线性混合整数规划模型,并设计基于遗传算法和粒子群算法的两阶段混合搜索求解算法.在此基础上,考虑开行列车在高速铁路网中的抗干扰能力,建立了面向网络化运营场景的开行方案综合评估指标体系,揭示了故障场景下高速铁路网络性能的演化规律.最后,以实际高速铁路线路数据和运营数据为场景进行仿真实验,本文提出方法在保证运输需求和路局收益的同时能够有效地提升8.66%网络整体负载均衡性,增强发生故障时网络的抗干扰能力.     

考虑天然来水随机性的水火电系统机组检修计划

如何合理安排机组检修是水火电系统调度运行中的一项重要任务。在长时间尺度下,天然来水的随机性使机组检修计划本质上成为随机优化问题,通常采用场景法描述随机性,但其形成的高维优化问题难以直接求解。建立多场景耦合的水火电系统机组检修优化模型,利用多学科协同优化(Multidisciplinary Collaborative Optimization, MCO)方法将各场景间的非预期性约束及检修变量耦合约束解耦,实现了原问题的降维,且MCO结构具有内在的并行性。此外,在基于MCO的系统级优化问题中,用绝对值惩罚项替代二次惩罚项,保证该问题是一个混合整数线性规划问题,有利于提高计算效率。最后以某省级实际水火电系统为算例进行仿真分析,验证了所提模型和算法的有效性。     

激光退火硅晶圆温度场分布的数值模拟研究

目前,激光退火技术被广泛应用于半导体加工领域,但对如何选择激光条件进行相应的退火并没有系统清晰的准则可以参考,尤其是在硅的深注入杂质激活方面。本文通过对激光照射在硅晶圆上形成的温度场分布进行数值模拟研究,分析了激光波长和脉冲宽度对加热深度以及晶圆背面温度的影响。结果表明,延长激光波长或脉冲宽度,都有助于增加激光退火的加热深度。而对于特定的激活深度需求,存在着最优的激光波长和脉冲宽度组合,可以使退火所需要的激光脉冲能量最低,硅晶圆背面的温升最小。本文通过模拟仿真给出了激活深度在1～10μm范围内的最优波长和脉宽值,可为实现高效的深硅注入激光激活工艺提供重要的条件参考。     

Er3+/Yb3+/Tm3+/Pr3+共掺碲酸盐玻璃超宽带近红外发光性能研究

采用熔融淬火技术制备了0.2 mol% Er₂O₃、1 mo l% Yb₂O₃、0.1 mol% Tm₂O₃和x mol% Pr₆O₁₁( x=0.25、 0.3、0.35和0.4)掺杂的65TeO₂-15ZnO-10Na₂O-10WO₃系碲酸盐玻璃,通过X 射线衍射(X-ray diffraction,XRD)、差热扫描曲线(differential scanning calorimetry,DSC)对玻璃样 品的抗析晶性和热稳定性 进行了表征。结果表明,玻璃样品具备良好的抗析晶性,析晶温度和转变温度差值为140 ℃,具有较好 的热稳定性。吸收光谱结果显示,Er³⁺/Yb³⁺/Tm³⁺/Pr³⁺共掺碲 酸盐玻璃在980 nm处有较强的吸收峰,故可 以采用980 nm泵浦源对该玻璃样品进行激发。在1 200—2 000 nm近 红外波段范围,玻璃样品存在峰值中 心为1.35 μm、1.53 μm和1.8 μm 3个波段发射峰,且3个发射峰的荧光半高宽(full width at half maxima, FWHM)均大于100 nm,其覆盖了光信号传输的E、S、C和C+L 4个波段,大幅度地提高了掺铒 光纤放大器(erbium doped fiber amplifier,EDFA)的放大带宽。     

基于策略学习的机票动态定价算法

机票动态定价旨在构建机票售价策略以最大化航班座位收益.现有机票定价算法都建立在提前预测各票价等级的需求量基础之上,会因票价等级需求量的预测偏差而降低模型性能.为此,提出基于策略学习的机票动态定价算法,其核心是不再预测各票价等级的需求量,而是将机票动态定价问题建模为离线强化学习问题.通过设计定价策略评估和策略更新的方式,从历史购票数据上学习具有最大期望收益的机票动态定价策略.同时设计了与现行定价策略和需求量预测方法的对比方法及评价指标.在两趟航班的多组定价结果表明:相比于现行机票销售策略,策略学习算法在座位收益上的提升率分别为30.94％和39.96％,且比基于需求量预测方法提升了6.04％和3.36％.     

一种准光模式变换器的设计和实验

准光模式变换器是实现高功率回旋管高效输出的重要部件。该文针对140 GHz, TE_28,8模回旋振荡管研制,开展以Denisov型辐射器和3个准光镜面构成的准光模式变换器设计与实验研究。利用标量衍射法优化辐射器辐射口径处的场分布,其与理想高斯场之间的矢量相关性为96.2%;基于几何光学和高斯波束匹配方法设计了聚焦镜面与波束整形镜面,采用3维全波分析软件Surf3D获得各个镜面上及输出窗处的场分布,对所设计的镜面系统进行了仿真验证,在输出窗处获得了高斯模式含量为96.67%的输出波束,整个模式变换器的功率转换效率为93.98%。以自行研制的TE_28,8模激励器作为准光模式变换器的输入,通过对模式变换器转换性能仿真结果验证,在严格控制加工精度及装配和实验过程的基础上,完成了准光模式变换器转换性能的冷测实验。实验结果表明,设计和实验具有合理的一致性,可以作为准光模式变换器工程应用设计和验证手段。     

Parallel Cloth Simulation Using OpenGL Shading Language

The primary goal of cloth simulation is to express object behavior in a realistic manner and achieve real-time performance by following the fundamental concept of physic. In general, the mass–spring system is applied to real-time cloth simulation with three types of springs. However, hard spring cloth simulation using the mass–spring system requires a small integration time-step in order to use a large stiffness coefficient. Furthermore, to obtain stable behavior, constraint enforcement is used instead of maintenance of the force of each spring. Constraint force computation involves a large sparse linear solving operation. Due to the large computation, we implement a cloth simulation using adaptive constraint activation and deactivation techniques that involve the mass–spring system and constraint enforcement method to prevent excessive elongation of cloth. At the same time, when the length of the spring is stretched or compressed over a defined threshold, adaptive constraint activation and deactivation method deactivates the spring and generate the implicit constraint. Traditional method that uses a serial process of the Central Processing Unit (CPU) to solve the system in every frame cannot handle the complex structure of clothmodel in real-time. Our simulation utilizes the Graphic Processing Unit (GPU) parallel processing with compute shader in OpenGL Shading Language (GLSL) to solve the system effectively. In this paper, we design and implement parallel method for cloth simulation, and experiment on the performance and behavior comparison of the mass–spring system, constraint enforcement, and adaptive constraint activation and deactivation techniques the using GPU-based parallel method.     

Re-weighting regression and sparsity regularization for multi-view classification

Applied Intelligence - Multi-view data widely exists in real life, which contains rich and comprehensive information. Multi-view learning aims to make full use of the information of multiple views...     

深度学习中的权重初始化方法研究

深度神经网络训练的实质是初始化权重不断调整的过程,整个训练过程存在耗费时间长、需要数据量大等问题。大量预训练网络由经过训练的权重数据组成,若能发现预训练网络权重分布规律,利用这些规律来初始化未训练网络,势必会减少网络训练时间。通过对AlexNet、ResNet18网络在ImageNet数据集上的预训练模型权重进行概率分布分析,发现该权重分布具备单侧幂律分布的特征,进而使用双对数拟合的方式进一步验证权重的单侧分布服从截断幂律分布的性质。基于该分布规律,结合防止过拟合的正则化思想提出一种标准化对称幂律分布（NSPL）的初始化方法,并基于AlexNet和ResNet32网络,与He初始化的正态分布、均匀分布两种方法在CIFAR10数据集上进行实验对比,结果表明,NSPL方法收敛速度优于正态分布、均匀分布两种初始化方法,且在ResNet32上取得了更高的精确度。     

Self-organizing maps for pattern recognition in design of alloys

A combined experimental–computational methodology for accelerated design of AlNiCo-type permanent magnetic alloys is presented with the objective of simultaneously extremizing several magnetic properties. Chemical concentrations of eight alloying elements were initially generated using a quasi-random number generator so as to achieve a uniform distribution in the design variable space. It was followed by manufacture and experimental evaluation of these alloys using an identical thermo-magnetic protocol. These experimental data were used to develop meta-models capable of directly relating the chemical composition with desired macroscopic properties of the alloys. These properties were simultaneously optimized to predict chemical compositions that result in improvement of properties. These data were further utilized to discover various correlations within the experimental dataset by using several concepts of artificial intelligence. In this work, an unsupervised neural network known as self-organizing maps was used to discover various patterns reported in the literature. These maps were also used to screen the composition of the next set of alloys to be manufactured and tested in the next iterative cycle. Several of these Pareto-optimized predictions out-performed the initial batch of alloys. This approach helps significantly reducing the time and the number of alloys needed in the alloy development process.