Investigation into impact of tropical cyclones on the ionosphere using GPS sounding and NCEP/NCAR Reanalysis data

Ionospheric response to tropical cyclones (TCs) was estimated experimentally on the example of three powerful cyclones – KATRINA (23–31 August 2005), RITA (18–26 September 2005), and WILMA (15–25 October 2005). These TCs were active near the USA Atlantic coast. Investigation was based on Total Electron Content (TEC) data from the international network of two-frequency ground-based GPS receivers and the NCEP/NCAR Reanalysis data. We studied the spatial–temporal dynamics of wave TEC disturbances over two periods of ranges (02–20 min and 20–60 min). To select the ionospheric disturbances which were most likely to be associated with the cyclones, maps of TEC disturbances were compared with those of meteorological parameters.     

Dynamic orbit propagation in a gravitational field of an inhomogeneous attractive body using the Lagrange coefficients

The analytical methods have nearly been replaced by the numerical methods due to their higher accuracy and accessibility of computation facilities. The semi-analytical Lagrange method of orbit propagation using f and g series is a competitive alternative to the numerical integration technique if the Lagrange coefficients are derived in a full gravitational field. In this paper, a generalization of the Lagrange method of orbit propagation is introduced. In other words, we introduce a complete form of the Lagrange coefficients in all force fields developed in the spherical harmonics for example full gravitational field of the Earth. The method is numerically compared with the numerical integration technique. In order to show the numerical performance of the method, it has been implemented for orbit propagation of a GPS-like MEO and CHAMP-like LEO satellites. Discrepancy at centimeter level for CHAMP-like and sub-millimeter accuracy for GPS-like satellites shows relatively high performance of the developed algorithm. Compared to integration method, the proposed Lagrange method is nearly faster by a factor two for small Nmax and four for large Nmax.     

发展性教学理论与网络环境下教学改革探讨

本文对赞科夫发展性教学理论的基本观点进行了分析,对发展性教学理论对我国教学改革的影响进行了概括,对发展性教学理论与网络环境下教学改革进行了探讨。     

某自动测试系统神经网络故障诊断模块设计与实现

阐述了自动测试系统中故障诊断的重要性,详述了诊断的流程.以某型防空导弹自动驾驶仪为例,首先利用LabVIEW中的MATLAB脚本节点,分别以BP和RBF神经网络为算法编写故障诊断函数;然后用LabVIEW开发故障诊断模块界面,最后将TabVIEW和神经网络两者的优势结合实现自动驾驶仪的故障诊断.实验仿真结果表明:模块设...     

载人飞船回收着陆半实物仿真系统关键技术研究

载人飞船回收着陆半实物仿真系统是通过模拟飞船返回压力环境,将回收程控装置硬件实物接入仿真回路的一个半实物仿真试验平台。文章简要介绍了该半实物仿真系统的结构,重点叙述了该半实物仿真系统中特有的负压高精度快速调节压力模拟、多通路指令信号处理、混合式网络、多阶段动力学仿真框架等关键技术。     

基于人工神经网络的柔性机翼挠度预测

为实现对承载后柔性机翼挠度的准确预测,在全面分析柔性机翼挠度的影响因素基础上,应用正交试验法确定的影响柔性机翼挠度的主要因子作为输入变量,挠度作为输出变量,以大量试验数据为训练样本,通过多次试取隐含层和各隐含单元,并选取trainlm作为最优训练函数,最终建立了预测柔性机翼挠度的BP(Back Propagation)人工神经网络模型.在此基础上,随机选取试验结果中的12组试验样本,连续进行10次挠度预测,预测结果和试验实测值最大相对误差和标准方差分别为4.481%,1.033 7.解析结果表明:柔性机翼挠度预测结果与实验值吻合的较好,建立的人工神经网络预测模型具有较高的预测精度.     

面向天基监视的红外弱小飞行目标识别算法

针对当前红外弱小飞行目标特征不明显、背景干扰大等问题,提出了一种基于深度学习的红外弱小目标识别算法。检测框架以YOLOv4模型为基础,通过使用K-means++算法对训练集的候选框进行聚类处理,在初始大小的选取上放弃随机生成初始点的方式,在样本集里选取某一个样本作为初始中心使锚框（anchor）大小的选取更加合理。在模型结构中引入卷积注意力模块,使算法模型计算资源分配更合理,对红外弱小飞行目标的特征信息更加敏感。改进空间金字塔池化模块,使用平均池化可以更多保留图像的原始信息,降低天基成像中的噪点与坏点的影响。仿真实验表明采用K-means++计算Anchor大小时准确率可以达到80.13%,在加入了SPP和CBAM模块后之后在测试集上算法识别准确率达到了83.3%,经过对模型的修改有效提升了对红外弱小飞行目标识别的准确率。     

基于BP神经网络的介质表面充电电位反演方法

为了实现全面、实时的在轨卫星充放电风险分析,基于在相同环境下,不同材料表面充电的关联性,利用BP神经网络建立了一种由Kapton材料表面充电电位反演卫星其他常用介质材料表面电位的模型。以Kapton材料的表面电位以及材料厚度为输入,其他介质材料的表面电位作为模型输出,使用COMSOL建立的表面充电模型对神经网络进行训练,将反演误差降低到10%以下,并利用Kapton与Teflon材料表面充电地面试验数据验证反演模型的准确性,结果显示Teflon表面电位的反演值与实测值间的相对误差小于16%。     

基于神经网络规则提取的航空发动机磨损故障诊断知识获取

针对神经网络智能诊断与专家系统中知识难于理解和诊断解释能力差等问题,研究了一种新的基于功能性观点的神经网络规则提取方法,介绍了方法流程及关键算法.并用UCI(加利福尼亚大学埃尔文分校)机器学习数据对方法进行了分析和验证.最后,将方法应用于实际航空发动机磨损故障诊断中,采集了某型航空发动机实测油样光谱数据237个样本,利用神经网络规则提取方法提取了发动机磨损故障诊断知识规则,并对其进行了解释,结果表明了方法的正确有效性.      

A Fast Global Node Selection Algorithm for Bearings-only Target Localization

In the target tracking, the nodes aggregate their observations of the directions of arrival of the target. The network then uses an extended Kalman filter （EKF） to combine the measurements from multiple snapshots to track the target. In order to rapidly select the best subset of nodes to localize the target with the minimum mean square position error and low power consumption, this paper proposes a simple algorithm, which uses the location information of the target and the network. The lower botmd of localization error is utilized according to the distances between the target and the selected active nodes. Furthermore, the direction likelihoods of the active nodes is predicted by way of the node/target bearing distributing relationships.