神经网络在导航卫星钟差预报中的应用CSCD

为提高导航卫星钟差预报精度,提出一种神经网络和多项式相结合的钟差预报方法,该方法在根据星载原子钟物理特性进行多项式模型预报后,采用神经网络对多项式模型预报误差进行建模,以实现导航卫星钟差预报精度补偿。为验证本文提出的预报模型的可行性和有效性,利用实测的COMPASS导航卫星钟差数据进行钟差预报精度分析,并与传统的多项式模型预报精度进行比较。结果表明:基于神经网络建立的组合预报模型能有效提高导航卫星钟差的预报精度。     

神经网络在导航卫星钟差预报中的应用

为提高导航卫星钟差预报精度,提出一种神经网络和多项式相结合的钟差预报方法,该方法在根据星载原子钟物理特性进行多项式模型预报后,采用神经网络对多项式模型预报误差进行建模,以实现导航卫星钟差预报精度补偿。为验证本文提出的预报模型的可行性和有效性,利用实测的COMPASS导航卫星钟差数据进行钟差预报精度分析,并与传统的多项式模型预报精度进行比较。结果表明:基于神经网络建立的组合预报模型能有效提高导航卫星钟差的预报精度。     

一种高精度实时GPS卫星钟差预测算法

为获得高精度实时GPS卫星钟差,文章提出一种基于多项式和最小二乘支持向量机(Least Squares Support Vector Machines,LS-SVM)相结合的钟差预报方法.该方法采用国际GNSS服务发布的超快速观测星历建模进行短期预报,首先根据卫星钟的物理特性用附有周期项的多项式模型进行拟合以提取趋势项和周期项,然后用LS-SVM对多项式拟合残差进行建模预报,最后将预报结果加上趋势项和周期项,得到最终的钟差预报值.试验结果表明,所提算法能够实时有效地对GPS卫星钟差进行预报,且精度优于超快速预报星历.     

基于深度学习递归神经网络的电离层总电子含量经验预报模型

利用行星际太阳风参数与太阳活动指数、地磁活动指数、电离层总电子含量格点化地图数据,首次基于一种能处理时间序列的深度学习递归神经网络（Recurrent Neural Network,RNN）,建立提前24h的单站电离层TEC预报模型.对北京站（40°N,115°E）的预测结果显示,RNN对扰动电离层的预测误差低于反向传播神经网络（Back Propagation Neural Network,BPNN）0.49～1.46TECU,将太阳风参数加入预报因子模型后对电离层正暴预测准确率的提升可达16.8%.RNN对2001和2015年31个强电离层暴预报的均方根误差比BPNN低0.2TECU,将太阳风参数加入RNN模型可使31个事件的平均预报误差降低0.36～0.47TECU.研究结果表明深度递归神经网络比BPNN更适用于电离层TEC的短期预报,且在预报因子中加入太阳风数据对电离层正暴的预报效果有明显改善.      

基于长短时记忆神经网络(LSTM)的钟差预报算法

原子钟钟差预报在原子时计算和原子钟频率驾驭中发挥着重要的作用。长短时记忆神经网络（LSTM）预报算法能够处理多参数长期依赖关系的时间序列预报,以氢钟和铯钟实测数据为样本,通过构建LSTM钟差预报模型,降低了长期原子钟内部噪声以及原子钟漂移对钟差预报的影响,并以72h,240h和720h为预报时长,分别与线性多项式模型、灰色模型和Kalman模型原子钟钟差预报模型进行预报误差对比。研究表明,在240h以上的预报时长中,LSTM建模长期依赖关系的优势得以体现,相较于其他3类模型可以获得更高的预报精度。     

一种高精度GPS卫星钟差预报方法

为了获得实时高精度GPS钟差,提出了采用快速星历建模进行短期预报。文章先对钟差数据提取趋势项,再利用傅里叶分析研究其周期特征以确定建模与预报时间段长度,最后利用径向基函数(Radial Basis Function,RBF)神经网络建模实时预报钟差。由于RBF神经网络用于非线性数据建模效果良好,在提取线性趋势项并合理确定建模周期后,该方法能够得到较好的预报结果。实际预报结果表明,文中方法得到的预报钟差精度高于超快速星历,能够满足分米级实时精密定位的要求。     

粒子群优化在直升机旋翼动平衡调整中的应用

传统的直升机旋翼调整方法没有考虑调整参数与振动信号之间的非线性关系,针对这一缺点,提出将广义回归神经网络(GRNN,General Regression Neural Network)和粒子群算法相结合的旋翼调整方法,采用GRNN网络建立旋翼动平衡调整模型,以桨叶的调整参数作为神经网络的输入,以旋翼转轴和机身的三向的加速度测量值作为网络输出,建立调整参数与直升机振动信号间的模型.以直升机振动作为目标函数,采用粒子群优化算法对桨叶的调整参数进行寻优,获得当直升机振动最小时的桨叶的调整量. 飞行实验结果表明,此方法可通过飞行测试获得的新数据对神经网络进行更新,使系统在使用过程中不断完善,并可在较少的飞行调整下完成旋翼的动平衡调整.      

灰色模型在卫星钟差预报中的缺陷分析

卫星钟差预报是时间同步的关键技术之一,方法繁多。针对灰色模型在卫星钟差预报中的应用,以GPS Rb钟钟差预报为实例,从模型构造上分析了该模型在卫星钟差预报中存在的缺陷,有利于该模型的改进和完善。     

基于灰色模型和ARIMA模型的钟差预测

由于多种因素的影响,原子钟运行情况十分复杂,为了有效地进行钟差预报和更好的反应原子钟的特性,提出一种基于灰色模型和ARIMA模型的组合模型来实现钟差的预报,并给出了灰色模型和ARIMA模型进行原子钟钟差预报的基本思想和预报模型。利用组合模型更精确地把握钟差序列复杂细致的变化规律,从而更好的逼近钟差序列。为了验证该组合模型的可行性和有效性,利用原子钟钟差数据进行钟差预报分析,结果显示该模型具有较好的预报精度。     

脉冲星方位误差估计的两步卡尔曼滤波算法

为了克服钟差和卫星位置误差对脉冲星方位误差估计的影响,设计了两步卡尔曼滤波（TSKF）算法。首先,介绍了脉冲星方位误差估计的传统模型,并通过分析和仿真验证了钟差、卫星位置误差以及2种误差同时存在时会使脉冲星方位误差估计结果产生较大偏差。其次,在传统的估计模型中加入了钟差和卫星位置误差,并将钟差和钟差变化率增广为新的状态量,从而推导出包含2种误差的新模型,并证明了该模型的完全可观测性;根据该模型并按照两步卡尔曼滤波原理,得到了TSKF算法的步骤。最后,通过仿真表明:在钟差和卫星位置误差同时影响下,传统脉冲星方位误差估计算法偏差较大且发散;TSKF算法则能够有效隔离2种误差的影响,使赤经和赤纬误差估计达到0.2 mas之内的精度。      

Forecasting magnetopause crossing locations by using Neural Networks

Given the highly complex and nonlinear nature of Near Earth Space processes, mathematical modeling of these processes is usually difficult or impossible. In such cases, modeling methods involving Artificial Intelligence may be employed. We demonstrate that data driven models, such as the Neural Network based approach, shows promise in its ability to forecast or predict the behavior of these processes. In this paper, modeling studies for forecasting magnetopause crossing locations are summarized and a Neural Network algorithm is presented to describe the nonlinear time-dependent response of the subsolar region of the magnetopause to varying solar wind conditions. In our approach the past history of the solar wind has, for the first time to the best knowledge of the authors, been included in forecasting the subsolar region of the magnetopause. It is proposed that the data driven approach is a valid approach to understanding and modeling the physical phenomena of Near Earth Space. The only basic requirement for the data driven approach is the availability of representative data for the phenomena. The objective of this paper is to demonstrate that by using WIND and GEOTAIL satellite data a Neural Network based model can be adapted to the modeling of the Earth’s magnetopause.     

广义回归神经网络在MC-CDMA多用户检测中的应用

在简要介绍了反传(BP)神经网络,径向基神经网络(RBFN),广义回归神经网络(GRNN)的基础上,可知GRNN与传统的基于反传(BP)算法的神经网络相比,有收敛速度快,鲁棒性强等优点,本文将广义回归神经网络(GRNN)技术援引到MC-CDMA多用户检测中,进行了误码率和收敛性的仿真计算,表明了GRNN在多用户检测应用中的优越性和有效性。     

Using TensorFlow-based Neural Network to estimate GNSS single frequency ionospheric delay (IONONet)

In the last 20?years, and in particular in the last decade, the availability of propagation data for GNSS has increased substantially. In this sense, the ionosphere has been sounded with a large number of receivers that provide an enormous amount of ionospheric data. Moreover, the maturity of the models has also been increased in the same period of time. As an example, IGS has ionospheric maps from GNSS data back to 1998, which would allow for the correlation of these data with other quantities relevant for the user and space weather (such as Solar Flux and Kp). These large datasets would account for almost half a billion points to be analyzed. With the advent and explosion of Big Data algorithms to analyze large databases and find correlations with different kinds of data, and the availability of open source code libraries (for example, the TensorFlow libraries from Google that are used in this paper), the possibility of merging these two worlds has been widely opened. In this paper, a proof of concept for a single frequency correction algorithm based in GNSS GIM vTEC and Fully Connected Neural Networks is provided. Different Neural Network architectures have been tested, including shallow (one hidden layer) and deep (up to five hidden layers) Neural Network models. The error in training data of such models ranges from 50% to 1% depending on the architecture used. Moreover, it is shown that by adjusting a Neural Network with data from 2005 to 2009 but tested with data from 2016 to 2017, Neural Network models could be suitable for the forecast of vTEC for single frequency users. The results indicate that this kind of model can be used in combination with the Galileo Signal-in-Space (SiS) NeQuick G parameters. This combination provides a broadcast model with equivalent performances to NeQuick G and better than GPS ICA for the years 2016 and 2017, showing a 3D position Root Mean Squared (RMS) error of approximately 2?m.     

测量时刻偏差对单星定轨误差的影响分析

为评估测量时刻偏差对单星定轨等效测量误差的影响,根据单星定轨处理策略分析了其理论模型,指出测站接收机的测量时刻偏差由测站时钟钟差以及测量时刻不准确度等组成。试验数据分析表明,测站钟差经一阶多项式拟合后的残差可近似为零均值的测量噪声;数值仿真结果表明,卫星信号发射时刻1ms误差导致GEO、IGSO、MEO三种卫星的等效测距误差分别为006cm、40cm、80cm。     

利用人工神经网络提前1h预报电离层TEC

提出了一种利用人工神经网络提前1h预报电离层TEC的简便方法. 考虑到实际工程应用要求, 没有使用其他空间天气参数, 而是选择电离层TEC观测数据本身作为输入参数. 输入参数为当前时刻TEC、一阶差分、相对差分和时间, 输出参数为预报时刻TEC. 利用文中介绍的GPS/TEC处理方法解算厦门站2004年电离层TEC观测数据, 对预报方法进行评估, 全年平均相对误差为9.3744%, 预报结果与观测值相关性达到了0.96678. 结果表明, 利用人工神经网络方法提前1h预报电离层TEC有很好的应用前景.      

利用神经网络预报电离层f0F2

由中国武汉电离层台站和澳大利亚Hobart台站的电离层F2层临界频率(f0F2)的资料,利用三层前向反馈神经网络(BP网络),提出一种提前24h预测f0F2的方法,该方法以前5天观测的f0F2数据拟合的5个系数以及太阳活动参数作为输入,以当天24 h的f0F2作为输出对网络进行训练,训练好的网络可以实现对f0F2提前24 h的预报.预测结果显示,利用神经网络预测的f0F2与实际观测结果变化趋势较一致,并且比IRI的计算结果更加准确.误差分析表明,在南半球Hobart(-42.9°,147.3°)台站比中国武汉站(30.4°,114.3°)的结果要好,在低年比高年要好,在冬夏季节比春秋季节稍好.本文说明利用神经网络对电离层参量进行预报是一种切实可行的方法.     

A new stochastic model considering satellite clock interpolation errors in precise point positioning

Precise clock products are typically interpolated based on the sampling interval of the observational data when they are used for in precise point positioning. However, due to the occurrence of white noise in atomic clocks, a residual component of such noise will inevitable reside within the observations when clock errors are interpolated, and such noise will affect the resolution of the positioning results. In this paper, which is based on a twenty-one-week analysis of the atomic clock noise characteristics of numerous satellites, a new stochastic observation model that considers satellite clock interpolation errors is proposed. First, the systematic error of each satellite in the IGR clock product was extracted using a wavelet de-noising method to obtain the empirical characteristics of atomic clock noise within each clock product. Then, based on those empirical characteristics, a stochastic observation model was structured that considered the satellite clock interpolation errors. Subsequently, the IGR and IGS clock products at different time intervals were used for experimental validation. A verification using 179 stations worldwide from the IGS showed that, compared with the conventional model, the convergence times using the stochastic model proposed in this study were respectively shortened by 4.8% and 4.0% when the IGR and IGS 300-s-interval clock products were used and by 19.1% and 19.4% when the 900-s-interval clock products were used. Furthermore, the disturbances during the initial phase of the calculation were also effectively improved.     

Application of neural networks to South African GPS TEC modelling

The propagation of radio signals in the Earth’s atmosphere is dominantly affected by the ionosphere due to its dispersive nature. Global Positioning System (GPS) data provides relevant information that leads to the derivation of total electron content (TEC) which can be considered as the ionosphere’s measure of ionisation. This paper presents part of a feasibility study for the development of a Neural Network (NN) based model for the prediction of South African GPS derived TEC. The South African GPS receiver network is operated and maintained by the Chief Directorate Surveys and Mapping (CDSM) in Cape Town, South Africa. Vertical total electron content (VTEC) was calculated for four GPS receiver stations using the Adjusted Spherical Harmonic (ASHA) model. Factors that influence TEC were then identified and used to derive input parameters for the NN. The well established factors used are seasonal variation, diurnal variation, solar activity and magnetic activity. Comparison of diurnal predicted TEC values from both the NN model and the International Reference Ionosphere (IRI-2001) with GPS TEC revealed that the IRI provides more accurate predictions than the NN model during the spring equinoxes. However, on average the NN model predicts GPS TEC more accurately than the IRI model over the GPS locations considered within South Africa.     

Analysis of GNSS clock prediction performance with different interrupt intervals and application to real-time kinematic precise point positioning

Continuous and timely real-time satellite orbit and clock products are mandatory for real-time precise point positioning (RT-PPP). Real-time high-precision satellite orbit and clock products should be predicted within a short time in case of communication delay or connection breakdown in practical applications. For prediction, historical data describing the characteristics of the real-time orbit and clock can be used as the basis for performing the prediction. When historical data are scarce, it is difficult for many existing models to perform precise predictions. In this paper, a linear regression model is used to predict clock products. Seven-day GeoForschungsZentrum (GFZ) final clock products sampled at 30 s are used to analyze the characteristics of GNSS clocks. It is shown that the linear regression model can be used as the prediction model for the satellite clock products. In addition, the accuracy of the clock prediction for different satellites are analyzed using historical data with different periods (such as 2 and 10 epochs). Experimental results show that the accuracy of the clock with the linear regression prediction model using historical data with 10 epochs is 1.0 ns within 900 s. This is higher accuracy than that achieved using historical data of 2 epochs. Finally, the performance analysis for real-time kinematic precise point positioning (PPP) is provided using GFZ final clock prediction results and state space representation (SSR) clock prediction results when communication delay or connection breakdown occur. Experimental results show that the positioning accuracy without prediction is better than that with prediction in general, whether using the final clock product or the SSR clock product. For the final clock product, the positioning accuracy in the north (N), east (E), and up (U) directions is better than 10.0 cm with all visible GNSS satellites with prediction. In comparison, the 3D positioning accuracy of N, E, and U directions with visible GNSS satellites whose prediction accuracy is better than 0.1 ns using historical data of 10 epochs is improved from 15.0 cm to 7.0 cm. For the SSR clock product, the positioning accuracy of N, E, and U directions is better than 12.0 cm with visible GNSS satellites with prediction. In comparison, the 3D positioning accuracy of N, E, and U directions with visible GNSS satellites whose prediction accuracy is better than 0.1 ns using historical data of 10 epochs is improved from 12.0 cm to 9.0 cm.