联邦学习安全与隐私保护综述

联邦学习是一种新型的分布式学习框架，它允许在多个参与者之间共享训练数据而不会泄露其数据隐私。但是这种新颖的学习机制仍然可能受到来自各种攻击者的前所未有的安全和隐私威胁。本文主要探讨联邦学习在安全和隐私方面面临的挑战。首先，本文介绍了联邦学习的基本概念和威胁模型，有助于理解其面临的攻击。其次，本文总结了由内部恶意实体发起的3种攻击类型，同时分析了联邦学习体系结构的安全漏洞和隐私漏洞。然后从差分隐私、同态密码系统和安全多方聚合等方面研究了目前最先进的防御方案。最后通过对这些解决方案的总结和比较，进一步讨论了该领域未来的发展方向。     

运载火箭推进剂复杂流动传热问题数值模拟中的模型简化方法

在运载火箭改进优化或新型运载火箭设计过程中,会遇到各种复杂的流动与传热问题。按照真实产品和物理过程直接建立的数值计算模型过于复杂,不同尺度的结构、流动与传热、长时间的飞行过程相互耦合,出现计算时间太长(数年)、收敛困难、程序调试困难等问题。针对初始条件复杂、飞行过程复杂、边界条件复杂3类具体问题,总结了3种相应的模型简化方法,分别是工程经验法、极限参数法、低维近似法,为推进剂复杂流动与传热的数值模拟提供参考。     

A multi-aircraft conflict detection and resolution method for 4-dimensional trajectory-based operation

Conflict Detection and Resolution(CDR) is the key to ensure aviation safety based on Trajectory Prediction(TP). Uncertainties that affect aircraft motions cause difficulty in an accurate prediction of the trajectory, especially in the context of four-dimensional(4D) Trajectory-Based Operation(4DTBO), which brings the uncertainty of pilot intent. This study draws on the idea of time geography, and turns the research focus of CDR from TP to an analysis of the aircraft reachable space constrained by 4D waypoint constraints. The concepts of space–time reachability of aircraft and space–time potential conflict space are proposed. A novel pre-CDR scheme for multiple aircraft is established. A key advantage of the scheme is that the uncertainty of pilot intent is accounted for via a Space-Time Prism(STP) for aircraft. Conflict detection is performed by verifying whether the STPs of aircraft intersect or not, and conflict resolution is performed by planning a conflict-free space–time trajectory avoiding intersection. Numerical examples are presented to validate the efficiency of the proposed scheme.     

A new interpretation of the historical records of observing Venus in daytime with naked eye: Focusing on the meteorological factors in the astronomical observation records

Observing Venus during the daytime is challenging. However, observational data of Venus during the daytime can be found in history books from East Asian countries such as Korea, China and Japan. In this study, we are focused on data from Korean history book, and using records left in the Joseon wangjo sillok, one of the Korean history books from the Joseon dynasty (CE 1392–1910), tries to prove whether the reported observations were actual observation data. We collected these 4663 records from the Joseon wangjo sillok to determine whether they were actual observations, and to confirm the scientific validity of the records. When we looked at the distribution of observations on a yearly basis, we noticed that there were more observations in the fall and winter, while there were only limited numbers of observations in the spring. This difference in distribution suggests that observations were strongly affected by atmospheric conditions due to the weather. From the 4663 data collected, we found a cycle of about 1.6?years. The cycle of about 1.6?years is assumed to be related to the conjunction cycle of Venus. And, we found a cycle of about 1.0?years. The cycle of about 1.0?years is assumed to reflect the indirect effects of the Asian dust phenomenon. Our research result verified the record as actual observational data. In addition, this result verified that meteorological factors were involved in the recorded astronomical observation data. We think that our data can be helpful to climate studies on the natural sources of key climate-affecting aerosols.     

应用后缘装置的跨声速层流翼型多岛优化设计

高雷诺数状态下,自然层流技术(Natural laminar flow,NLF)是减小机翼表面湍流摩擦阻力的有效方法。然而由于层流翼面上大范围顺压梯度的存在使得后缘处的恢复压差更大,产生更强的激波。因此在减小摩擦阻力的同时又增加了激波阻力。本文采用后缘装置(Trailing edge device,TED)来控制翼型后缘处的激波强度,基于线性稳定性理论(Linear stability theory,LST)的eN方法对流动进行转捩判断,进而应用多岛并行多目标进化算法(multi-objective evolutionary algorithm,MOEA)以获得大范围层流区域和弱化激波强度为目标对翼型进行优化设计。优化结果表明合作均衡策略耦合进化算法可以快速地捕捉到该多目标问题的Pareto阵面解,阵面上翼型的波阻力和摩擦阻力性能较初始翼型大大改善。同时,采用后缘装置控制激波强度时,无论在设计点还是偏离设计点时,优化后翼型均具有良好的升阻力特性和鲁棒性。     

类IXV飞行器初期再入制导与姿态控制方法研究

摘要： 针对类IXV飞行器无翼式升力体构型及采用RCS/尾襟翼组合控制特点，研究其初期再入段的高精度制导与姿态控制方法.设计带过程约束的数值预测 校正制导律以提高制导系统的鲁棒性及精确性；根据其执行机构配置特点，设计基于RCS的航向控制律及基于RCS/气动舵的纵、横向复合控制律，并采用鲁棒伺服LQR技术进行控制参数快速设计.通过蒙特卡洛打靶仿真来验证算法的精度及鲁棒性.     

涡轮基组合循环发动机分布式控制系统通信网络混合拓扑结构优化方法

涡轮机组合循环(Turbine based combined cycle,TBCC)发动机控制系统通信网络拓扑结构是其分布式控制系统方案设计的重要部分,优化网络拓扑结构可提高发动机推重比和控制系统可靠性。本文基于智能优化算法提出TBCC分布式控制系统网络拓扑结构优化方法。基于图论建立TBCC几何模型和网格模型,以重量和可靠性为优化性能指标,同时考虑发动机表面高温区域以及控制节点的工作可靠性,分别采用粒子群算法和遗传算法优化星形结构中智能中央节点位置、中央节点的环形拓扑结构,获得星形-环形混合拓扑结构。仿真实例表明,基于本文方法优化所得的混合拓扑结构相较于星形集中式控制结构,系统重量降低了51.9%。     

多臂空间机器人的视觉伺服与协调控制

针对多臂空间机器人自主目标抓捕任务，首先建立多臂空间机器人的运动模型和其与目标的相对运动模型，采用Kane方法建立多臂空间机器人的动力学模型；其次，研究基于视觉伺服的机械臂在线轨迹规划算法，并引入零反作用机动，消除机械臂运动对平台姿态的扰动；再次，在不使用零反作用机动功能时，分别使用基于角动量前馈补偿的协调控制算法和逆动力学方法设计了协调控制器，在机械臂运动时保持平台姿态和相对目标的位置。最后，开发了基于Matlab的仿真软件MASS（多臂空间机器人仿真），仿真结果校验了上述方法的有效性。     

新能源接入配电网经济性与可靠性优化研究

为寻求新能源配电网的可靠性和经济性最优,实现新能源电源的优化配置,本文提出将可靠性和经济性同时作为两个优化目标,运用自适应粒子群优化法,对新能源电源进行多目标优化配置。以IEEE-RBTS Bus 6主馈线F4为例,通过系统仿真找到新能源配电系统可靠性与经济较均衡的Pareto最优解集,可根据决策者的偏好选择最优方案。本文的研究为新能源电力系统可靠性和经济评估提供了一个新的研究思路。     

飞机动态推出控制策略仿真优化模型

机场离港运行的无序化造成滑行过程中的长时间排队等待及大量燃油浪费。为了减少燃油消耗和废气排放,在已有的动态推出控制策略基础上提出了阶梯函数控制策略(Step function,SPC)和非线性函数动态推出策略(Nonlinear function,NPC)的一般形式,以离港成本为目标,建立了基于停机位等待惩罚的动态推出控制模型,在不延误的前提下提出了一种基于网格参数优化的蒙特卡洛仿真优化算法。通过北京首都机场实际运行数据对推出过程进行仿真计算,并与无控制策略以及传统N-control策略的最优可达解进行仿真对比,结果表明:在不延误的前提下,提出的推出策略可以更加有效地降低平均滑行道滑行时间,NPC策略的离港运行成本和燃油成本可降低45.52%和54.23%,SPC虽然成本节省劣于NPC策略,但是其简单的操作方式可以为离港推出调度方式的改进提供决策支持。