Agent仿真中具有先验知识的混合学习算法与混合结构模型

强化学习是一种有效的机器学习方法,是无监督学习,通过不断地和环境交互得到外部环境评价信号,选择合适的动作。Q学习是一种典型的强化学习,其学习效率较低,尤其是当状态空间和决策空间较大时。为提高Q学习学习效率和收敛速度,采用具有先验知识的Q学习算法,利用模糊综合决策方法处理专家经验和环境信息得到Q学习的先验知识,对Q学习的初始状态进行优化;针对Agent个体学习与群体学习各自的不足,提出了采用混合学习算法,将个体学习与群体学习有效结合起来,提高了Agent的个体性能及系统整体的智能水平;同时为满足复杂适应性需求,采用Agent混合结构模型,在该模型中构造了基于知识的协调控制器,通过它来协调慎思式过程和反应式过程。     

改进的图像相关匹配算法

将遗传算法用于低对比度小目标图像相关匹配跟踪,提出了改进的快速图像相关匹配算法。结合图像匹配本身的特点,采用新的编码方式克服了二进制编码的缺点。采用NPROD匹配准则在同样的匹配效果条件下,提高了匹配精度,该算法在计算速度上较原始算法提高了至少12倍以上,同时该算法还具有每帧图像匹配计算时间基本恒定的优点,便于在实际系统中采用。实验结果表明该方法可以有效地跟踪低对比度小目标。     

FADS系统迭代算法的收敛性分析

嵌入式大气数据传感系统采用分布在飞行器前端周线上的压力传感器阵列测量的压力推导大气参数的.介绍了采用迭代法推导大气参数的方法,针对迭代过程中出现的发散问题,提出采用小扰动线性化分析法解决多变量叠代算法的收敛性分析.最后应用Matlab仿真验证,仿真结果表明,在一般的超声速状态下(马赫数小于4.5),该算法是收敛的.     

大规模混合网格的分区策略研究

在飞行器阻力数值模拟中常常需要数百万网格点的巨型混合网格,甚至需要网格点数达到数千万的超巨型混合网格.在并行计算前往往需要对这些大规模混合网格进行分区操作.针对Metis库函数所需要输入的大规模边表的生成问题,提出了一种生成边表的并行算法,并运用此算法,在微机机群上对具有4,787,893个网格点的DLR-F6翼型(翼身组合体 挂架 发动机短舱)的巨型混合网格成功地生成21,556,110条边的大规模边表,进而采用Metis提供的多级循环二分法顺利地对该巨型混合网格实现均匀分区,从而验证了该并行算法的有效性.     

一种新的基于位置信息的路由算法研究

 为航空移动Ad hoc网络(MANET)提出一种基于位置信息的路由算法,即位置网格路由（LBGR）算法,以此来解决节点高速移动引起的路径重建问题。算法中,数据包沿路由发现时得到的网格轨迹进行传输,每个收到数据包的中转节点根据它与相邻节点、目的节点的位置关系及网络轨迹下游的路由网格位置决定下一跳节点。由于利用了网格位置信息,LBGR算法在航空节点高速移动和拓扑变化无法事先预测的情况下,使用较少的开销就可获得稳定的路由。仿真结果表明,与GRID路由算法相比,LBGR算法在节点中高速移动时,能够有效地减少路由开销,提高网络吞吐量。     

微小卫星剩磁在轨标定技术研究

 基于微小卫星姿态确定系统常采用无陀螺配置方案,克服环境干扰力矩的影响并提高微小卫星姿态确定的精度是此类姿态确定系统的关键。分析了星上稳定剩余磁场对无陀螺微小卫星姿态确定的干扰机理,建立了卫星剩余磁矩与磁强计偏置标定模型。以磁强计、太阳敏感器作为姿态敏感器件,并采用扩展卡尔曼滤波器实现标定算法,为无陀螺磁测微小卫星消除剩磁干扰,获得高精度姿态估计提供了一种新方法。仿真结果表明：该方法能准确估计卫星剩余磁矩与磁强计偏置,磁强计偏置的标定精度在1 nT左右,剩余磁矩的标定精度为0.000 1 A·m²量级,有效消除了剩磁对无陀螺卫星姿态确定的影响,显著提高了姿态确定精度,滤波器能在500 s内收敛。     

Optimization and Sizing for Propulsion System of Liquid Rocket Using Genetic Algorithm

Flight vehicle conceptual design appears to be a promising area for application of the Genetic Algorithm (GA) as an approach to help to automate part of the design process. This computational research effort strives to develop a propulsion system design strategy for liquid rocket to optimize take-off mass, satisfying the mission range under the constraint of axial overload. The method by which this process is accomplished by using GA as optimizer is outlined in this paper. Convergence of GA is improved by introducing initial population based on Design of Experiments Technique.     

Dynamic Response Analysis of Bird Strike on Aircraft Windshield Based on Damage-modified Nonlinear Viscoelastic Constitutive Relation

Damage-modified nonlinear viscoelastic constitutive equation and failure criterion are introduced and the three-dimensional incremental forms are deduced based on the updated Lagrangian approach. A simple tensile test model and a split Hopkinson pressure bar model are built to verify the accuracy of the subroutine implemented within the non-linear finite element program LS-DYNA. A numerical model of bird strike on windshield is established to study the responses of windshield under three different bird velocities at three sites. The bird is represented by a cylinder with a hemisphere at each end and the contact-impact coupling algorithm is used in this study. It is found that the implemented subroutine can properly describe the mechanical behavior of polymethyl methacrylate under low and high strain rates and large deformation, and can be used validly.     

Interactive Multi-objective Optimization Design for the Pylon Structure of an Airplane

The pylon structure of an airplane is very complex, and its high-fidelity analysis is quite time-consuming. If posterior preference optimization algorithm is used to solve this problem, the huge time consumption will be unacceptable in engineering practice due to the large amount of evaluation needed for the algorithm. So, a new interactive optimization algorithm-interactive multi-objective particle swarm optimization （IMOPSO） is presented. IMOPSO is efficient, simple and operable. The decision-maker can expediently determine the accurate preference in IMOPSO. IMOPSO is used to perform the pylon structure optimization design of an airplane, and a satisfactory design is achieved after only 12 generations of IMOPSO evolutions. Compared with original design, the maximum displacement of the satisfactory design is reduced, and the mass of the satisfactory design is decreased for 22%.