广义回归神经网络在日长变化预报中的应用

传统的日长变化预报多是基于线性模型,如最小二乘模型、自回归模型等,但是日长变化包含了复杂的非线性因素,线性模型预报的效果往往不甚理想.所以尝试使用一种非线性神经网络—广义回归神经网络(GRNN)模型进行日长变化预报,并将结果与使用BP (Back Propagation)神经网络模型和其它模型的预报结果进行比较.结果表明,GRNN用于日长变化预报是高效可行的.     

两种样本输入方式下基于GRNN的日长变化预报结果的比较

针对广义回归神经网络用于日长变化预报过程中,样本的输入方式对预报结果的影响进行了研究。采用2种输入方式:即样本按不同跨度输入以及按连续输入,对日长变化进行预报。最终证明不同的样本输入方式对日长变化预报精度的影响较大,样本按跨度输入在超短期预报中预报精度较高,样本采用连续输入的方式在短期和中期预报中预报精度较高。     

遗传算法优化的BP神经网络卫星钟差预报

针对BP (Back Propagation)神经网络模型预测卫星钟差中权值和阈值的最优化问题, 提出了基于遗传算法优化的BP神经网络卫星钟差短期预报模型, 给出了遗传算法优化BP神经网络的基本思想、具体方法和实施步骤. 为验证该优化模型的有效性和可行性, 利用北斗卫星导航系统(BeiDou navigation satellite system, BDS)卫星钟差数据进行钟差预报精度分析, 并将其与灰色模型(GM(1,1))和BP神经网络模型预报的结果比较分析. 结果表明: 该模型在短期钟差预报中具有较好的精度, 优于GM(1,1)模型和BP神经网络模型.     

Application of artificial neural networks to the prediction of dust storms in Northwest China

Artificial neural networks (ANN) are non-linear mapping structures analogous to the functioning of the human brain. In this study, we take the ANN approach to model and predict the occurrence of dust storms in Northwest China, by using a combination of daily mean meteorological measurements and dust storm occurrence. The performance of the ANN model in simulating dust storm occurrences is compared with a stepwise regression model. The correlation coefficients between the observed and the estimated dust storm occurrences obtained from the neural network procedure are found to be significantly higher than those obtained from the regression model with the same input data. The prediction tests show that the ANN models used in this study have the potential of forecasting dust storm occurrence in Northwest China by using conventional meteorological variables.     

The Important Role Played by Atmospheric Angular Momentum in the Non-linear Prediction of the Variations in the Length of Day

The variation in the length of day has complicated time-varying characteristics and the traditional method for linear time series analysis is always difficult to obtain good effect of prediction. If the non-linear artificial neural network technique is adopted to predict the variation in the length of day, the topological structure of the network model is determined by the least square error method. By taking into account the close relation between the variation in the length of day and the general circulation of atmosphere, the axial sequence of atmospheric angular momentum is introduced into the forecasting model of neural network. The results show that the forecast accuracy is significantly improved by taking advantage of the combination of the length of day and the atmospheric angular momentum sequence in comparison with the individual adoption of the data of the length of day.     

Short-term prediction of fof2 using time-delay neural network

To test the ability and efficacy of neural networks in short-term prediction of ionospheric parameters, this study used the time series of the ionospheric foF2 data from Slough station during solar cycles 21 and 22. It describes different neural network architectures that led to similar conclusions on one-hour- ahead foF2 prediction. This prediction is compared with observations and results from linear and persistence models considered here as two special cases of the neural networks.     

基于深度学习的太阳F_10.7辐射通量的短期预报研究

F_10.7太阳辐射通量作为输入参数被广泛运用于大气经验模型、电离层模型等空间环境模型,其预报精度直接影响航天器轨道预报精度.采用时间序列法统计了太阳辐射通量F10.7指数和太阳黑子数(SSN)的关系,给出了两者之间的线性关系,在此基础上提出了一种基于长短时记忆神经网络(Long and Short Term Memory,LSTM)的预报方法,方法结合了54 d太阳辐射通量指数和SSN历史数据来对F_10.7进行未来7 d短期预报,并与其他预报方法的预报结果进行了比较,结果表明:(1)所建短期预报7 d方法模型的性能优于美国空间天气预报中心(Space Weather Prediction Center, SWPC)的方法,预测值和观测值的相关系数(CC)达到0.96,同时其均方根误差约为11.62个太阳辐射通量单位(sfu),预报结果的均方根误差(RMSE)低于SWPC,下降约11%;(2)对预测的23、24周太阳活动年结果统计表明,太阳活动高年的第7 d F10.7指数预报平均绝对百分比误差(MAPE)最优可达12.9%以内,低年最优可达2...     

基于深度学习的LAMOST星系星族参数测量

星系的光谱包含其内部恒星的年龄和金属丰度等信息, 从观测光谱数据中测量这些信息对于深入了解星系的形成和演化至关重要. LAMOST (Large Sky Area Multi-Object Fiber Spectroscopic Telescope)巡天发布了大量的星系光谱, 这些高维光谱与它们的物理参数之间存在着高度的非线性关系. 而深度学习适合于处理多维、海量的非线性数据, 因此基于深度学习技术构建了一个8个卷积层$+$4个池化层$+$1个全连接层的卷积神经网络, 对LAMOST Data Release 7 (DR7)星系的年龄和金属丰度进行自动估计. 实验结果表明, 使用卷积神经网络通过星系光谱预测的星族参数与传统方法基本一致, 误差在0.18dex以内, 并且随着光谱信噪比的增大, 预测误差越来越小. 实验还对比了卷积神经网络与随机森林回归模型、深度神经网络的参数测量结果, 结果表明卷积神经网络的结果优于其他两种回归模型.     

The real-space power spectrum of the PSCz survey from 0.01 to 300 h Mpc−1

We report a measurement of the real-space (not redshift-space) power spectrum of galaxies over four and a half decades of wavenumber, 0.01 to 300  h  Mpc⁻¹, from the IRAS Point Source Catalog Redshift Survey (PSC z ). Since estimates of power are highly correlated in the non-linear regime, we also report results for the pre-whitened power spectrum, which is less correlated. The inferred bias between optically selected APM and IRAS -selected PSC z galaxies is about 1.15 at linear scales ≲0.3  h  Mpc⁻¹, increasing to about 1.4 at non-linear scales ≳1  h  Mpc⁻¹. The non-linear power spectrum of PSC z shows a near power-law behaviour to the smallest scales measured, with possible mild upward curvature in the broad vicinity of   k ∼2  h  Mpc⁻¹  . Contrary to the prediction of unbiased dark matter models, there is no prominent inflection at the linear to non-linear transition scale, and no turnover at the transition to the virialized regime. The non-linear power spectrum of PSC z requires scale-dependent bias: all Dark Matter models without scale-dependent bias are ruled out with high confidence.     

Evolution of peaks in weakly non-linear density field and dark halo profiles

Using the two-point Edgeworth series up to second order in the linear rms density fluctuation we construct the weakly non-linear conditional probability distribution function for the density field around an overdense region. This requires calculating the two-point analogues of the skewness parameter S ₃. We test the dependence of the two-point skewness on distance from the peak for scale-free power spectra and Gaussian smoothing. The statistical features of such a conditional distribution are given as the values obtained within linear theory corrected by the terms that arise as a result of weakly non-linear evolution. The expected density around the peak is found to be always below the linear prediction while its dispersion is always larger than in the linear case. For large enough overdensities the weakly non-linear corrections can be more significant than the peak constraint introduced by Bardeen et al. We apply these results to the spherical model of collapse as developed by Hoffman & Shaham and find that in general the effect of weakly non-linear interactions is to decrease the scale from which a peak gathers mass and therefore also the mass itself. In the case of an open universe this results in steepening of the final profile of the virialized proto-object.     

Prediction of Clock Errors of Atomic Clocks Based on Modified Linear Combination Model†

The prediction of the clock errors of atomic clocks plays an important role in the work on time and frequency. Each of the prediction models often used at present has its own merits and shortages. A combination of the predicted results obtained by means of these models can be used to synthesize the characteristics of various kinds of prediction models. In the light of the problem which occurs when the linear combination model is used to make the prediction of clock errors, the concept of learning weight is proposed and the modified combination prediction model is made by taking advantage of various kinds of pieces of accuracy information. For verifying the efficiency of this method the clock error sequences of the IGS (International GNSS Service) of 4 GPS satellites are selected and the predicted results of the quadratic polynomial and grey model are combined. The result shows that the modified model can further improve the stability and accuracy based on the guarantee of the reliability.     

原子钟信号的神经网络模型

时间尺度的连续性要求对原子钟信号进行必要的预测,预测的实质是建立一个模型来逼近原子钟的信号,前向神经网络具有良好的副近非线性 函数的能力,用神经网络模型来预测原子钟信号,并与AD模型的预测结果作了比较。     

On the application of information theory to the optimum state-space reconstruction of the short-term solar radio flux (10.7 cm), and its prediction via a neural network

The prediction of a time series using a neural network involves an optimum state-space reconstruction. The state space of the daily 10.7-cm solar radio flux is reconstructed using an information theory approach. A multi-layer feed-forward neural net is used for short-term prediction of the time series. The convergence of the synaptic weights is obtained partially by simulated annealing and partially by the 'quick prop' variation of back-propagation. The result gives a reasonably accurate short-term prediction.     

Real time Kp predictions from solar wind data using neural networks

Multilayer feed-forward neural network models are developed to make three-hour predictions of the planetary magnetospheric Kp index. The input parameters for the networks are the B_z-component of the interplanetary magnetic field, the solar wind density n, and the solar wind velocity V, given as three-hour averages. The networks are trained with the error back-propagation algorithm on data sequences extracted from the 21^st solar cycle. The result is a hybrid model consisting of two expert networks providing Kp predictions with an RMS error of 0.96 and a correlation of 0.76 in reference to the measured Kp values. This result can be compared with the linear correlation between V(t) and Kp(t + 3 hours) which is 0.47. The hybrid model is tested on geomagnetic storm events extracted from the 22^nd solar cycle. The hybrid model is implemented and real time predictions of the planetary magnetospheric Kp index are available at http://www.astro.lu. se/-fredrikb.     

Quinstant dark energy predictions for structure formation

We explore the predictions of a class of dark energy models, quinstant dark energy, concerning the structure formation in the Universe, both in the linear and non-linear regimes. Quinstant dark energy is considered to be formed by quintessence and a negative cosmological constant. We conclude that these models give good predictions for structure formation in the linear regime, but fail to do so in the non-linear one, for redshifts larger than one.     

Atmospheric Excitation of Time Variable Length-of-Day on Seasonal Scales

We use the method of wavelet transform to analyze the time series of the Earth's rotation rate of the EOP(IERS)C04.The result shows that the seasonal (annual and semiannual)variation of the length-of-day(LOD)has temporal vari- ability in its period length and amplitude.During 1965.0-2001.0,the periods of the semiannual and annual components varied mainly from 175-day to 190-day and from 360-day to 370-day,respectively;while their amplitudes varied by more than 0.2 ms and 0.1 ms,respectively.Analyzing the axial component of atmospheric angular mo- mentum(AAM)during this period,we have found that time-variations of period lengths and amplitudes also exist in the seasonal oscillations of the axial AAM and are in good consistency with those of the seasonal LOD change.The time variation of the axial AAM can explain largely the change of the LOD on seasonal scales.     

Non-linear redshift distortions: the two-point correlation function

We consider a situation where the density and peculiar velocities in real space are linear, and we calculate ξ _s , the two-point correlation function in redshift space, incorporating all non-linear effects which arise as a consequence of the map from real to redshift space. Our result is non-perturbative and it includes the effects of possible multi-streaming in redshift space. We find that the deviations from the predictions of the linear redshift distortion analysis increase for the higher spherical harmonics of ξ _s . While the deviations are insignificant for the monopole ξ ₀, the hexadecapole ξ ₄ exhibits large deviations from the linear predictions. For a COBE normalized     ,     cold dark matter (CDM) power spectrum, our results for ξ ₄ deviate from the linear predictions by a factor of two on the scale of ∼10  h ⁻¹ Mpc. The deviations from the linear predictions depend separately on f (Ω) and b . This holds the possibility of removing the degeneracy that exists between these two parameters in the linear analysis of redshift surveys which yields only     .
We also show that the commonly used phenomenological model, where the non-linear redshift two-point correlation function is calculated by convolving the linear redshift correlation function with an isotropic pair velocity distribution function, is a limiting case of our result.     

基于线性数据重构的天线动力学模型辨识方法

射电望远镜天线伺服控制系统中的非线性特性, 对系统动力学特性辨识有着显著的影响, 会提高辨识难度, 增加辨识模型的复杂程度. 系统非线性特性的测量与补偿也会增加系统辨识工作量. 针对上述问题, 提出了一种基于非线性采样数据的线性重构方法, 用于动力学特性建模. 通过提取原采样数据的相位与幅值, 对受到噪声与非线性畸变影响的系统采样数据进行线性重构, 降低待辨识模型的复杂度. 搭建了半实物实验平台, 以平台实际采样为基础, 重构线性数据, 利用奇异值法与自回归神经网络评估并辨识平台动力学模型. 实验结果表明, 建模数据奇异值拐点从100阶下降至40阶, 仅用10个神经网络节点200次训练即实现了模型辨识.