基于BP神经网络的纳卫星轨道温度预测

为了实现纳卫星在轨温度的预测,在对纳卫星热系统动态特性模型分析的基础上,建立BP神经网络预测模型实现纳卫星在轨温度的预测.通过分析纳卫星热系统动态特性模型,得到用于BP神经网络预测模型的输入、输出变量以及训练神经网络所需的数据样本.BP神经网络预测模型分别以纳卫星外壳、辐射器、舱内仪器的热流及温度值为神经网络输入、输出,预测纳卫星10s后的轨道温度.经验证,神经网络预测模型预测结果与纳卫星实际轨道温度吻合较好,表明神经网络预测模型是快捷有效的.      

利用BP神经网络预测AE指数

采用的预报模式是一种全连接的BP网络模型,利用太阳风及行星际磁场的观测数据预报AE指数.神经网络输入选用ACE卫星数据,取5 min平均值,通过比较,选用4个预报参量.构造了预报参量时续为20 min,40 min和60 min依次递增的三个网络,分别进行训练和预测,并对行星际参量对AE指数影响的时续性进行了探讨.预报结果表明,全连接BP神经网络在AE指数的短期预报中是比较有效的,同时还提出了需要进一步改进的环节.      

应用BP神经网络针对高通量质子事件的通量数值预报

质子事件的爆发与太阳软X射线辐射有着很强的相关性,利用GOES卫星的1～8 (A)波段和0.5～4 (A)波段的软X射线数据,选取一些特征参量验证该相关性并应用到质子事件短期预报中.在当前质子事件传输物理机制不完全明确的情况下,在现有的预报质子事件有无的模型基础上,利用BP神经网络,根据软X射线通量水平等预测事件质子峰值通量水平,再对训练后的网络进行检验,检验预测所得结果与实际探测值误差小于一个量级,具备一定实用意义.      

IRI-2007对扩展F的预测与观测比较研究

通过对赤道地区两台站的扩展F 实际观测值与IRI-2007 模型预测值的比较, 研究了该模型预测结果的地域局限性和预测的准确度. 该模型建立在美洲扇区赤道附近台站资料基础上, 本文选取非洲扇区大体同纬度两个台站资料进行对比研究. 统计结果表明, 无论在发生概率日均值的年变化趋势上, 还是在扩展F 发生率年均值随地方时变化以及随太阳活动的统计中, 都存在着明显的差异; 而对非洲两站实测值的分析和IRI-2007 模型对南美巴西地区的预测又分别与当前已有的研究结果相符. 分析结果表明, (1) 建立在巴西台站数据基础上的IRI-2007 模型的统计建模极具参考价值; (2) 赤道和低纬度电离层扩展F 现象有强烈的经度效应, 在该模型建模方法基础上, 结合不同扇区实际资料的统计, 能改进该模型全球应用的普适性; (3) IRI-2007模型能正确反映巴西地区扩展F 的发生概率, 用该模型和其他扇区不同条件下的实测值对比, 提供了一个细致研究扩展F经度效应的基础, 有利于确认导致扩展F发生的基本条件和多种影响因素的具体作用.      

利用人工神经网络快速计算木星系磁坐标

在木星辐射带研究中，从地理坐标向磁坐标的准确转换是建模基础.以往的建模中，磁壳参数L值的计算基于磁偶极场假设，该方法精确度较差.结合最新的高精度木星磁场模型JRM09，本文提出基于磁力线追踪法的木星磁坐标计算方法，并分析其合理性和必要性.要求精确度较高时，磁力线追踪法计算耗时很长.本文在磁力线追踪法的基础上进行改进，提出基于人工神经网络的磁坐标快速计算方法.该方法包括分类器和拟合器.分类器基于Adaboost算法的BP神经网络，用于预测某地理坐标是否在内磁层，如果在内磁层，则用拟合器计算L值.拟合器采用遗传算法优化BP神经网络.结果表明，分类器的分类错误率在3%以内，而拟合器的预测误差在7%以内.以Juno号一圈探测轨道为例，利用神经网络的磁坐标计算法比磁力线追踪计算法速度快3个数量级以上.基于人工神经网络的磁坐标快速计算方法可用于未来木星辐射带的研究.      

一种电离层TEC格点预测模型

基于分析时间序列数据的门限控制单元（GRU）神经网络模型，利用电离层TEC网格点历史数据、太阳活动指数、地磁活动指数作为预测因子，提出一种高精度电离层TEC格点预测模型.对全球60个网格点的数据进行了模型预测和对比实验，得到北半球平均相对精度的均值为83.96%，高于南半球的73.60%，表明预测模型在北半球的适应性更好，且中低纬地区的适应性优于高纬地区；预测模型在磁扰动期的平均相对精度的均值比磁平静期平均相对精度的均值高，约1.95%；与基于递归神经网络（RNN）、长短时记忆网络（LSTM）和双向长短时记忆网络（Bi-LSTM）的电离层TEC单站预测模型相比，本文预测模型的均方根误差（RMSE）平均为原来的80.8%.      

NRLMSISE-00大气模型与GRACE和CHAMP卫星大气密度数据的对比分析

利用GRACE(Gravity Recovery And Climate Experiment)和CHAMP(Challenging Mini-Satellite Payload)卫星2002-2008年的大气密度数据与NRLMSISE-00大气模型密度结果进行比较,分析了模型密度误差及其特点.结果显示,NRLMSISE-00大气模型计算的密度值普遍偏大,其相对误差随经纬度变化,在高纬度相对较小;相对误差随地方时变化,在02:00LT和15:00LT左右较大,10:00LT和20:00LT左右较小.通过模型密度相对误差与太阳F_10.7指数的对比分析发现,在太阳活动低年模型相对误差最大,而在太阳活动高年相对误差较小;将模型结果分别与GRACEA/B双星和CHAMP卫星的密度数据进行比较,发现对于轨道高度更高的GRACE卫星轨道,模型相对误差更大;在地磁平静期,相对误差与地磁ap指数(当前3h)相关性不强,但是在大磁暴发生时,误差急剧增大.      

积冰几何特性对翼型性能影响的神经网络预测

积冰几何形状对翼型气动系数的影响是复杂的.采用BP(Back Propagation)神经网络的LM(Levenberg-Marguardt)学习算法,建立明冰的典型几何特性(冰角前缘半径、冰角高度和冰角位置)对翼型气动系数影响的神经网络,得到该3种几何参数对气动系数影响的规律;建立了典型冰形参数对最大升力系数影响的神经网络,该网络能很好的预测冰形参数对应的最大升力系数值;此外,建立了冰型位置对舵面铰链力矩系数影响的神经网络.仿真结果表明,BP神经网络仿真结果与实验值具有高度一致性,并能预测非实验值条件下的气动系数;翼型表面积冰位置变化对气动系数影响最大;铰链力矩系数在失速迎角达到之前就发生突变,可以更安全地用来预测失速的发生.      

IRI-2007预测TEC在广州地区的适用性分析

利用广州站(23.2°N, 113.3°E) GPS双频接收机监测的电离层TEC数据和IRI-2007模型不同电离层输入参数计算得到的TEC预测值, 对比分析了太阳活动低年(2008年)广州地区TEC的变化特征. 结果表明, TEC观测值周日变化在16:00LT左右达到最大值, 而IRI-TEC最大值出现时间较GPS-TEC提前1h左右. TEC季节变化在春秋分较高, 两至季节较低, 表现出明显的半年特性和季节依赖性, 并出现冬季异常现象. IRI-TEC与GPS-TEC在白天具有较好的一致性, 夜间偏差较大. 不同电离层输入参数得到的TEC预测值也相差较大, 选用顶部电子密度参数NeQuick、底部厚度参数B0 Table并用URSI系数计算F₂层峰值参数时, 能较好地反映TEC观测值的变化特征. 在对磁暴的响应上, 预测值无明显变化, 观测值则有比较明显的表现. 通过对比, 初步分析了利用IRI-2007模型预测TEC在广州地区的适用性, 并给出了合理的参数选择方案.      

基于LightGBM的卫星飞轮温度预测

为保障卫星的正常在轨运行，地面系统需要对卫星运行状态进行监控预警，其中对卫星各系统的温度监控尤为重要.温度不仅直接反映卫星系统的健康状态，更会对系统器件的性能和寿命造成影响.飞轮作为卫星姿态控制系统的重要组件，其温度变化是识别姿态控制系统状态的重要信息.卫星飞轮温度的预测与预警对卫星在轨稳定运行具有重要意义.本文基于某在轨卫星遥测数据，结合空间环境数据，应用LightGBM机器学习框架研究建立梯度提升决策树模型，对卫星飞轮温度进行预测.经与实际遥测温度值进行对比验证，预测精度可以满足对卫星飞轮温度的监视需求.研究结果可应用于地面系统，对卫星姿态控制系统可能发生的温度异常进行预警，使地面运控人员能够提前规避风险，保障卫星的安全在轨运行.      

Neural network prediction of the topside electron content over the Euro-African sector derived from Swarm-A measurements

This study presents the first prediction results of a neural network model for the vertical total electron content of the topside ionosphere based on Swarm-A measurements. The model was trained on 5 years of Swarm-A data over the Euro-African sector spanning the period 1 January 2014 to 31 December 2018. The Swarm-A data was combined with solar and geomagnetic indices to train the NN model. The Swarm-A data of 1 January to 30 September 2019 was used to test the performance of the neural network. The data was divided into two main categories: most quiet and most disturbed days of each month. Each category was subdivided into two sub-categories according to the Swarm-A trajectory i.e. whether it was ascending or descending in order to accommodate the change in local time when the satellite traverses the poles. Four pairs of neural network models were implemented, the first of each pair having one hidden layer, and the second of each pair having two hidden layers, for the following cases: 1) quiet day-ascending, 2) quiet day-descending, 3) disturbed day-ascending, and 4) disturbed day-descending. The topside vertical total electron content predicted by the neural network models compared well with the measurements by Swarm-A. The model that performed best was the one hidden layer model in the case of quiet days for descending trajectories, with RMSE = 1.20 TECU, R = 0.76. The worst performance occurred during the disturbed descending trajectories where the one hidden layer model had the worst RMSE = 2.12 TECU, (R = 0.54), and the two hidden layer model had the worst correlation coefficient R = 0.47 (RMSE = 1.57).In all cases, the neural network models performed better than the IRI2016 model in predicting the topside total electron content. The NN models presented here is the first such attempt at comparing NN models for the topside VTEC based on Swarm-A measurements.     

Ionospheric TEC prediction using hybrid method based on ensemble empirical mode decomposition (EEMD) and long short-term memory (LSTM) deep learning model over India

Total electron data (TEC) from GPS nowadays can be used as a tool for understanding the space weather phenomena. The development of prediction model for TEC is quiet crucial and challenging due to the dynamic behavior of the ionosphere, since it depends on different factors such as seasonal, diurnal and spatial variations, solar geomagnetic conditions etc. In this paper, an attempt is made for predicting the GPS derived TEC values for different GNSS stations over India using a hybrid method based on Ensemble empirical mode decomposition (EEMD) and Long Short-Term Memory (LSTM) deep learning method. The daily TEC time series data from the IISc Bangalore (Latitude 13.021, Longitude 77.570), Lucknow (Latitude 26.912, Longitude 80.956) and Hyderabad (Latitude 17.417, Longitude 78.551) stations over India during the period 2008 to 2015 of solar cycle 23 and 24 is used for analysis. The assessment of model performance for testing predicted output compared with LSTM and EMD-LSTM models, and their comparison results show that the hybrid EEMD-LSTM model presents better than the other models.     

The equatorial ionization anomaly at the topside F region of the ionosphere along 75°E

Electron density measured by the Indian satellite SROSS C2 at the altitude of ∼500 km in the 75°E longitude sector for the ascending half of the solar cycle 22 from 1995 to 1999 are used to study the position and density of the equatorial ionization anomaly (EIA). Results show that the latitudinal position and peak electron density of the EIA crest and crest to trough ratios of the anomaly during the 10:00–14:00 LT period vary with season and from one year to another. Both EIA crest position and density are found to be asymmetric about the magnetic equator and the asymmetry depends on season as well as the year of observation, i.e., solar activity. The latitudinal position of the crest of the EIA and the crest density bears good positive correlation with F_10.7 and the strength of the equatorial electrojet (EEJ).     

利用人工神经网络预报大磁暴

本文采用阈值预报的策略和人工神经网络BP模型,以13个太阳风参量和地磁AE,Dst指数作为输入,以0或1作为输出,提前4h预报大磁暴主相发生的时刻．结果表明,采用神经网络方法的阈值预报可以对灾害性磁暴的发生提前数小时做出比较准确的预报．     

Modeling M(3000)F2 based on extreme learning machine

This paper presents an novel extreme learning machine (ELM)-based prediction model for the ionospheric propagation factor M(3000)F2 at Darwin station (12.4°S, 131.5°E; −44.5°dip) in Australia. The proposed ELM model is trained with hourly daily values of M(3000)F2 from the period 1998–2014 except 2001 and 2009. The hourly daily values of 2001 (high solar activity) and 2009 (low solar activity) are used for validating the prediction accuracy. The proposed ELM for modeling M(3000)F2 can achieve faster training process and similar testing accuracy compared with backward propagation neural network (BPNN). In addition, the performance of the ELM is verified by comparing the predicted values of M(3000)F2 with observed values and the international reference ionosphere (IRI −2016) model predicted values. Based on the error differences (the root mean square error (RMSE) and the M(3000)F2 percentage improvement values M(3000)F2_IMP(%)), the result demonstrates the effectiveness of the ELM model compared with the IRI-2016 model at hourly, daily, monthly, and yearly in high (2001) and low (2009) solar activity years. The ELM also shows good agreement with observations compared with the IRI during disturbed magnetic activity.     

Model of the slot region of earth''s electron radiation belt depending on the heliospheric parameters

Relativistic electrons in the slot region of Earth's electron radiation belt are studied using CORONAS-I satellite data obtained in March–May 1994. The strong shifts of the slot latitudinal location (from L≈3 to L≈2) were found. These shifts are associated with the Earth crossing sector boundaries formed by sector corotating structures of the solar wind and interplanetary magnetic field (IMF). The quantitative analysis of the relationship between properties of the slot region and values of the solar wind and IMF parameters was undertaken. The empirical model of the slot region dynamics was developed by means of Artificial Neural Network (ANN).     

低轨航天器弹道系数估算及热层大气模型误差分析

利用低轨（LEO）航天器在轨期间两行轨道根数（TLEs）数据,结合经验大气密度模型NRLMSISE00,反演计算得到其在轨期间的弹道系数B’,以31年B’的平均值代替弹道系数真值,分别通过标准球形目标卫星对比以及物理参数基本相同的非球形目标卫星对比,对弹道系数真值进行了检验;利用不同外形目标卫星弹道系数在不同太阳活动周内的变化规律,结合太阳和地磁活动变化,估计经验大气密度模型的误差分布. 结果表明,利用反演弹道系数31年的平均值来代替真值,其在理论值的正常误差范围内;大气密度模型误差在210～526km高度范围内存在相同的变化趋势,且模型误差随高度增加而增大;在短周期内B’变化与太阳活动指数F_10.7存在反相关性;密度模型不能有效模拟2008年出现的大气密度异常低. 以上结果表明,经验大气密度模型结果需要修正,尤其是在太阳活动峰年和谷年,此外,磁暴期间模型误差的修正对卫星定轨和轨道预报等也具有重要意义.