陕西省北斗三号区域电离层TEC建模与精度评估

基于陕西省北斗地基增强监测站数据,利用4阶球谐函数建立了陕西区域电离层总电子含量(TEC)模型,生成区域TEC格网产品.实验结果表明：以欧洲定轨中心(CODE)全球电离层图(GIM)产品作为参考,利用BDS-3数据获得的陕西区域北斗电离层产品精度为2.54 TECu,与CODE产品标称精度相当.在单频精密单点定位(SF-PPP)性能方面,BDS-3区域电离层产品可以提升定位精度35%,优于CODE全球电离层产品改正效果(27%).综合而言,BDS-3单系统在区域电离层建模方面具备较为优异的性能水平.     

用于风电功率预测的RPCL优化神经网络模型

为了提高风电功率预测的准确度,提出了一种基于对手竞争惩罚学习算法( rival penalized competitive learning,RPCL)优化径向基函数( radial basis function,RBF)神经网络的风电功率预测模型。首先通过RPCL确定网络隐含层神经元数目以及中心点初始值,然后由K均值聚类法确定隐含层神经元的中心点和宽度,最后通过最小均值算法确定隐含层神经元与输出层神经元之间的权值。仿真结果表明：此优化模型相较于传统RBF网络具有更高的准确性。     

The variation of the estimated GPS instrumental bias and its possible connection with ionospheric variability

Based on the analysis of the satellite DCB data estimated by our method and the Center for Orbit Determination in Europe(CODE)from 1999 to 2011,the features of the temporal variation of differential code biases(DCB)are studied.Summarily,there are three types of variations in DCB on different time scales.The first one is the day-to-day variation that exhibits more obviously in solar maximum years.The second one is the variation with about one year periodic variation that behaves more obviously from 1999 to 2004.The last one is the monotonously descending tendency from 1999 to 2010.Considering the basic ionospheric approximation in DCB estimation method,the features of the variability of the ionospheric morphology from 1999to 2010 are also displayed based on the ionospheric characteristic parameters.It can be concluded that the day-to-day and annual variation of the estimated global positioning system(GPS).DCB is related to the ionospheric variability.The variation of DCBs on solar cycle time scale includes the real hardware DCBs and pseudo-DCBs induced by ionospheric variation.No doubt,these kinds of"pseudo"variations of DCB will affect the precision of ionospheric total electron content(TEC)derived from the GPS data.In addition,this study is helpful for evaluating the influence of ionospheric weather on TEC derivation and is also useful for developing one estimation method of DCB with more stability and precision through introducing a more practical ionospheric model.     

DE算法改进的炼焦能耗RBF预测模型

针对炼焦能耗计算繁琐、影响因素众多的问题,以目标火道温度、烟道吸力、水分、挥发分和炼焦时间为输入变量,以炼焦能耗为输出变量,提出基于差分进化算法改进的RBF预测模型。由于RBF网络存在学习能力差、收敛速度慢等多个缺点,针对性地改进了差分进化算法优化的能耗预测模型。利用具有强大全局搜索能力的差分进化算法,选择RBF网络中基函数的中心值、宽度和输出权重的计算最优值,以此作为RBF神经网络的中心值、宽度和输出权重。结果表明,改进后的RBF预测模型具有较高的精度、稳定性和训练速度,对降低炼焦能耗、提高焦炭产量和提高企业经济效益具有重要意义。     

CSES GNSS ionospheric inversion technique,validation and error analysis

With the increased number of low Earth orbit (LEO) satellites equipped with global navigation satellite system (GNSS) receiver, the LEO based GNSS slant total electron content (TEC) and electron density profile (EDP) data play an increasingly important role in space weather and ionospheric research due to improved global coverage. China Seismo-Electromagnetic Satellite (CSES), which was launched in February 2018, is equipped with GNSS receiver for either precise orbit determination (POD) and ionospheric inversion. The purpose of the present paper is to validate CSES GNSS ionospheric inversion technique based on the real observations and verify the accuracy of TEC and EDP retrieval based on the simulated data. The following conclusions can be drawn: the epoch difference inversion (EDI) derived from CSES can successfully retrieve the EDPs without non-occultation side measurements; the technique of EDI and the calibrated TEC inversion (CTI) have similar behaviors in inversion errors, however, the retrieved NmF2 and hmF2 have a larger systematic error surrounding the equatorial ionization anomaly (EIA) where the assumption of spherical symmetry is often invalid; the precision and accuracy of retrieved TEC have been investigated in the paper based on the simulated data, and it is found that the accuracy of the retrieved TEC is relative to solar activity: the lower the F10.7 index, the higher the accuracy of retrieved TEC.     

UWSNs节点运动模型及预测定位

为提高水下无线传感器网络(UWSNs)中动态节点的定位精度,降低通信能耗,提出采用节点的运动模型实现预测定位.考虑到近海监控网络中,潮汐运动是海水运动的主要成因,以粗略的近海潮汐运动模型为基础,以高斯径向基函数作为空间基函数构造节点的运动模型;利用K-medoids方法对模型中的高斯径向基函数中心进行聚类优化;提出了采用扩展卡尔曼滤波的方法实现模型系数的估计.考虑到普通节点与锚节点运动的空间相关性,设计了与到锚节点的距离相关的权重系数,以锚节点的运动模型系数估计普通节点运动模型中的系数,进而完成自身定位.对东经117.25°—132.2°,北纬24°—43.45°海域UWSNs的节点定位性能进行仿真分析,结果表明:所提出的节点预测定位方法的定位性能较高,定位覆盖度和定位精度高于SLMP方法和MP-PSO方法,平均通信能耗低于SLMP方法和MP-PSO方法.所提出的节点预测定位方法适用于大规模水下动态无线传感器网络定位.     

大地震前震中上空电离层全电子含量异常研究

采用IGS(International GPS Service)网站提供的电离层垂直全电子含量(vertical total elearon content,记为VTEC),分析研究了1998—2006年发生在中国大陆地区的30个M_s≥6.0的地震.采用统计处理的方法,讨论了地震前后15d内电离层全电子含量的变化,分别对震中上空电离层全部TEC资料进行了分析.     

Global ionospheric TEC response to a strong magnetic storm

The global characteristics of the ionospheric storm and irregularities as well as propagation of TEC (total electron content) disturbances during the strong magnetic storm occurring in November 2004 were investigated by using the data of the IGS network. For the response of the global ionospheric TEC to this strong magnetic storm, the following features are noticeable: 1) the maximum of the ionospheric storm phase occurred around the main phase maximum of the magnetic storm; 2) the TEC response in equatorial and low latitudes was more remarkable than that in mid-high latitudes; 3) as a whole, the storm phase in the northern hemisphere was mainly positive, and it was negative in the southern hemisphere; 4) during the whole magnetic storm from November 7 to 11, the locations where the maxima of the positive and negative ionospheric storm phases occurred were nearly invariant to the Sun at low and equatorial latitudes, i.e. the 24-h recurrence. Analyzing results of TEC rate and its standard deviation showed that the ionospheric irregularities and disturbances in the global mainly occurred around the main phase maximum of the storm, and they distributed in a large longitudinal region for both day and night in mid-high latitudes and they generated and developed only after the sunset, and lasted out to the midnight in equatorial and low latitudes. The disturbance propagation parameters were also estimated by using the wavelet reconstruction and cross-correlation technologies for a set of spaced stations in the Northern America. Supported by the National Natural Science Foundation of China (Grant Nos. 40474055, 40504019)     

自适应BAS优化RBF神经网络的短时交通流预测

为提高短时交通流预测精度,针对传统径向基函数(radial basis function, RBF)神经网络短时交通流预测模型中心值固定、易受漂移数据干扰问题,提出自适应天牛须搜索算法(beetle antennae search algorithm, BAS)优化RBF神经网络的短时交通流预测模型。模型采用自适应步长提高BAS算法迭代速度和寻优能力,结合DBSCAN聚类确定RBF神经网络隐含层径向基函数网络中心,进而优化神经网络结构。通过路网真实交通流数据进行训练,选择常用于短时交通流预测的BP神经网络,RBF神经网络,广义RBF神经网络进行对比。结果表明：优化后的模型预测结果相较BP神经网络平均绝对误差降低了1.87%、平均绝对百分比误差降低了15.96%、均方根误差降低了3.24%,拟合度提高了3.96%;相较广义RBF神经网络平均绝对误差降低1.36%、平均绝对百分比误差降低了5.01%、均方根误差降低了2.19%,拟合度提高了2.5%。改进后的短时交通流预测模型能够为智能交通诱导提供可靠的预测值。     

基于RBF -BP神经网络融合的医学数据分类研究

为改善传统的反向传播(BP)神经网络和径向基函数(RBF)神经网络的学习能力和分类性能的不足,提出一种融合RBF网络与BP网络的混合神经网络算法(HRBF -BP),并将其应用到医学数据分类问题中.在网络结构的实现上,将RBF隐藏层与BP隐藏层进行级联融合,即在连接BP网络输入层与隐藏层之间加入RBF核映射层; 在学习算法的实现上,先采用k-均值聚类算法来实现RBF核参数的估计,然后再使用基于随机梯度下降的BP算法实现级联BP网络的权值优化.将该算法与SGBP、KMRB、PFRBF等算法在不同的医学数据集上进行分类实验对比表明,该方法的网络训练精度以及测试精度均优于SGBP、KMRB、PFRBF算法; 因此,该方法对提高BP网络和RBF网络的学习能力和分类性能具有良好的参考价值.     

基于势函数聚类的改进RBF网络算法研究

为优化径向基函数(RBF)网络结构并改善网络泛化性能,提出了一种基于势函数聚类的改进RBF网络算法.首先使用势函数统计每个模式类别中的样本势值,以此实现样本空间中不同样本的势值度量; 其次以增量学习的方式逐次完成对样本空间的有效覆盖,以此实现网络隐节点个数及参数的自动有效估计.最后将本文算法与KMRBF、FCRBF、MRAN以及 GAP -RBF 学习算法进行了实验对比,结果表明本文算法的网络分类精度更高,克服了KMRBF和FCRBF算法需人工调整网络隐节点来提高分类精度的问题,且比 GAP -RBF和MRAN算法的网络结构更加简单.     

GA优化的RBF神经网络外骨骼灵敏度放大控制

为改善外骨骼机器人灵敏度放大控制(SAC)性能,结合遗传算法(GA)与径向基函数(RBF)神经网络建立在线计算外骨骼机器人的精确动力学模型.用GA优化RBF神经网络的中心矢量与基宽度,并对RBF网络的权值实时更新,在线学习外骨骼机器人动力学模型中的参数矩阵,进一步推导出SAC控制律.仿真结果表明:GA优化后的RBF网络,可以在线学习外骨骼的动力学模型,基于该模型的SAC能够实现精确的人体轨迹跟踪,相比于优化前,人体轨迹跟踪误差以及人机交互信息会快速减小并收敛到0的微小邻域内,可实现人机协调运动.     

基于GA-RBF神经网络的电力系统短期负荷预测

针对短期负荷预测问题,提出了一种遗传算法-径向基函数(GA-RBF)神经网络负荷预测方法,解决传统径向基函数(RBF)神经网络预测中难以确定最佳隐藏层数问题,以提高预测的准确性。首先分析了GA算法模型和RBF神经网络模型;然后利用GA算法与RBF模型结合得到GA-RBF负荷预测模型;最后利用仿真工具对所建模型进行训练和预测。结果表明,与传统方法相比,其平均绝对百分误差值降低了4. 7%,证明了该方法的精确性和有效性。     

RBF神经网络的一种快速鲁棒学习算法

研究RBF神经网络的一种具有对训练数据集中离群点的鲁棒性的快速学习算法。使用相减聚类(Subtractive Clustering,SC)法选择RBF网络隐结点的中心,以定标鲁棒损失函数(Scaled Robust Loss Function,SRLF)为目标函数,采用梯度下降法调整隐结点的宽度和网络权值,从而使RBF网络的学习过程不受离群点的影响,并且能够快速收敛。实验结果表明了RBF神经网络的这一学习算法的优越性。     

基于RBF网络上界自适应学习的预警卫星滑模控制

分析了RBF(径向基函数)神经网络的基本结构和数学特性,对于预警卫星动力学系统的不确定性上界值无法测量和未知的情况,采用RBF神经网络可以对较强干扰上界进行自适应学习,并可降低控制和动力学带来的抖振。针对带有摆镜的预警卫星姿态控制问题,提出了一种基于神经网络扰动补偿的姿态滑模控制方法。针对RBF网络正交最小二乘(OLS)学习算法,采用RBF神经网络来学习不确定因素的上界值,并设计了预警卫星的姿态控制规律,解决了预警卫星动力学扰动补偿问题。利用数值仿真估算了基于RBF网络上界自适应学习滑模控制的预警卫星姿态控制系统的性能指标。     

Radial Basis Function Neural Network Based Super-Resolution Restoration for an Undersampled Image

To achieve restoration of high frequency information for an undersampled and degraded low-resolution image, a nonlinear and real-time processing method the radial basis function (RBF) neural network based super-resolution method of restoration is proposed. The RBF network configuration and processing method is suitable for a high resolution restoration from an undersampled low-resolution image. The soft-competition learning scheme based on the k-means algorithm is used, and can achieve higher mapping approximation accuracy without increase in the network size. Experiments showed that the proposed algorithm can achieve a super-resolution restored image from an undersampled and degraded low-resolution image, and requires a shorter training time when compared with the mulfiplayer perception (MLP) network.     

基于SOM-RBF神经网络的用电量预测模型研究

随着我国电网建设的高速发展,从日常电力负荷变化趋势剖析未来年度用电量已经成为电网建设的关键问题之一。根据1997~2016年湖北省年用电量及其10个影响因子的数据作为样本,提出了一种自组织特征映射神经网络(Self-organizing Feature Maps,SOM)与多变量的径向基函数(Radial Basis Function,RBF)结合的人工神经网络预测模型新方法。采用先聚类、再分类预测的方法,解决了由于RBF神经网络对于少量样本和训练样本点分散所导致的预测精度降低的问题,改进的神经网络泛化能力有所提高。结果表明:通过SOM-RBF组合算法进行预测,其相对误差维持在3%以下,平均相对误差为1.88%,预测效果较BP神经网络和RBF神经网络有较大的提升。这表明SOM-RBF组合算法可有效的用于用电量预测,具有较高的实用价值。     

Radial Basis Function Neural Network Based Super-Resolution Restoration for an Underspled Image

To achieve restoration of high frequency information for an underspled and degraded low-resolution image, a nonlinear and real-time processing method-the radial basis function (RBF) neural network based super-resolution method of restoration is proposed. The RBF network configuration and processing method is suitable for a high resolution restoration from an underspled low-resolution image. The soft-competition learning scheme based on the k-means algorithm is used, and can achieve higher mapping approximation accuracy without increase in the network size. Experiments showed that the proposed algorithm can achieve a super-resolution restored image from an underspled and degraded low-resolution image, and requires a shorter training time when compared with the multiplayer perception (MLP) network.     

RBF神经网络在电力负荷预测中的应用

将ＲＢＦ（ＲａｄｉａＢａｓｉｓＦｕｎｃｔｉｏ．辐基函数）人工神经网络模型用于电力系统日峰值负荷与日小时负荷的预测。文中首先给出了ＲＢＦ网络的结构，然后讨论确定ＲＢＦ网络中心及网络训练的聚类法和正文化法。利用从京津唐系统中收集到的负荷数据进行网络模型的训练和回响检测，所得结果证实了ＡＢＦ网络用于负荷预测的有效性。     

基于OCA客观聚类的RBF神经网络的研究

影响RBF神经网络性能的关键因素是基函数中心的选取,而目前尚没有可靠的方法选取RBF神经网络的中心。基于GMDH理论的OCA客观聚类具有能够自动确定最优聚类个数的优点。将OCA聚类应用于RBF神经网络中,用以自适应确定隐节点数目和各径向基函数中心,克服了传统RBF网络不能客观确定隐节点数目的缺点。实验仿真结果表明,基于OCA客观聚类的RBF神经网络具有自适应性、正确率高和训练速度快的优点。