基于强化学习的在轨目标逼近

从强化学习的角度,对在轨目标逼近问题进行研究,设计了一种整合制导与控制的端到端的算法.首先对在轨目标逼近问题进行数学建模;然后对强化学习算法原理进行简介,根据问题特点分析不同强化学习框架的优劣,确定以DDPG作为算法框架,并设计了基于强化学习的在轨目标逼近算法;最后通过仿真验证,分析了基于强化学习逼近算法的优劣性.     

高温声表面波器件Pt电极的制备研究*

针对高温声表面波(SAW)器件中的Pt电极制备需求,采用电感耦合等离子体干法刻蚀工艺实现Pt电极的干法刻蚀。通过采用纯Ar气源,研究不同ICP功率/RF功率下Pt电极的刻蚀,采用优化的刻蚀参数实现SAW中Pt叉指电极的制备。针对Pt溅射刻蚀中出现的再沉积问题,分析沉积物对电极制备的影响,通过后处理实现沉积物的去除,实现高温SAW器件的制备。     

Research on navigation of satellite constellation based on an asynchronous observation model using X-ray pulsar

Pulsar navigation is a promising navigation method for high-altitude orbit space tasks or deep space exploration. At present, an important reason for restricting the development of pulsar navigation is that navigation accuracy is not high due to the slow update of the measurements. In order to improve the accuracy of pulsar navigation, an asynchronous observation model which can improve the update rate of the measurements is proposed on the basis of satellite constellation which has a broad space for development because of its visibility and reliability. The simulation results show that the asynchronous observation model improves the positioning accuracy by 31.48% and velocity accuracy by 24.75% than that of the synchronous observation model. With the new Doppler effects compensation method in the asynchronous observation model proposed in this paper, the positioning accuracy is improved by 32.27%, and the velocity accuracy is improved by 34.07% than that of the traditional method. The simulation results show that without considering the clock error will result in a filtering divergence.     

A New D-GCL for Unidirectional Motion with Large Displacement

Numerical simulations of unsteady flow problems with moving boundaries commonly require the use of geometric conservation law(GCL).However,in cases of unidirectional large mesh deformation,the cumulative error caused by the discrete procedure in GCL can significantly increase,and a direct consequence is that the calculated cell volume may become negative.To control the cumulative error,a new discrete GCL(D-GCL)is proposed.Unlike the original D-GCL,the proposed method uses the control volume analytically evaluated according to the grid motion at the time level n,instead of using the calculated value from the D-GCL itself.Error analysis indicates that the truncation error of the numerical scheme is guaranteed to be the same order as that obtained from the original D-GCL,while the accumulated error is greatly reduced.For validation,two challenging large deformation cases including a rotating circular cylinder case and a descending GAW-(1)two-element airfoil case are selected to be investigated.Good agreements are found between the calculated results and some other literature data,demonstrating the feasibility of the proposed D-GCL for unidirectional motions with large displacements.     

微纳聚合体卫星的对称式重构规划算法

为解决微纳聚合体卫星变构过程中，聚合体各部分之间的动力学耦合导致的卫星整体姿态翻转或紊乱，提出了一种对称式变构规划算法。首先建立了铰链约束下的漂浮基多刚体系统动力学模型，研究了重构过程中各运动模块对本体姿态的影响，提出当每两个运动模块位置与转动方向满足对称性条件时，两者对本体姿态的影响可在一定程度上相互抵消。基于上述理论，在A*算法中引入最优分配度量和对称性判定，设计了并行对称重构规划算法。仿真结果表明，该规划算法可实现重构过程中多模块并行、对称运动，重构过程总步数较少，运动模块对本体姿态的影响小。     

高校新媒体环境下大学生意识形态教育新路径

意识形态关乎社会稳定和国家前途,必须得到足够的重视。而意识形态教育中青年学生的问题又尤为突出,当代大学生思想活跃,自我意识强烈,对意识形态领域的教育较为反感,因此意识形态教育工作必须创新其内容和形式。高校新媒体平台因为其新颖性和碎片化的特点,可以很好的作为意识形态教育工作的载体,承担意识形态教育工作的重任。     

基于深度学习的组合体航天器模型预测控制

利用模型预测算法先预测控制结果后控制的类人行为特点，借助深度学习在多参数寻优上的优势，提出了一种基于卷积神经网络的模型预测控制算法，满足航天工程低硬件需求，实现组合航天器多场景下姿态控制律的重构。该算法首先利用模型预测控制将组合航天器从初始状态控制到预期状态，然后将控制过程中状态量用于3层3核卷积神经网络的训练，训练完成后，用该卷积神经网络代替模型预测对组合航天器进行控制，从而降低计算资源需求。仿真校验表明：该算法可预测5个控制周期内的控制参数，相比传统模型预测算法所需硬件计算时间降低约5倍，在一般硬件环境下30 s内即可完成各场景下的组合航天器姿态控制，控制精度在10 -4 量级。     

空间低温杜瓦瓶凯夫拉支撑结构多约束优化设计

采用新型凯夫拉纤维作为某空间用小型低温杜瓦瓶的支撑材料，在验证其力学性能的基础上，设计了杜瓦瓶的支撑结构，建立了一种新的热力学耦合优化模型，给出支撑结构固有频率、热应力及发射环境下的约束条件，通过一种“离散点加密”新约束优化法计算出了支撑结构满足约束条件范围内的要求的最优设计参数。