基于误差最小的SVM最优分类面修正

针对C-支持向量机(C-SVM,C-Support Vector Machine)中惩罚系数C可能导致最优分类面不合理的问题,提出基于误差最小的SVM最优分类面修正方法.通过调整正负类分类间隔的约束条件,求解使训练样本总误差最小的偏置系数,并兼顾与正负类误差之差的绝对值的平衡,得到误差最小的更优分类面.实验证明该修正方法与C-SVM及其它修正方法相比,具有较高的分类精度和较强的抗噪声与野值数据干扰能力.     

SVM最优分类面相对位置的修正

通过放宽标准支持向量机(SVM,Support Vector Machines)中类别边界至分类面等间隔的约束,保持两类函数间隔之和不变的条件,在支持向量机思想框架下得出分类面依样本分布进行调整的新型支持向量机,其对偶形式与标准支持向量机完全相同,从而在理论上进一步完善了支持向量机.在此基础上,提出使类别的函数间隔正比于样本标准差的具体算法--方差修正法,达到最优分类面的相对位置依样本方差而调整之目的.从统计意义上来说,方差修正法在分类精度上有所提高,但计算量增加不多.     

基于贝叶斯准则的支持向量机预测模型

对实际统计数据中存在的相关性、不确定性和非线性问题,提出贝叶斯支持向量机预测模型方法.构建基于高斯分布的权值分布模型描述信息的不确定性,基于先验概率分布和贝叶斯关系获得后验分布模型,利用极大似然方法和递推迭代算法求解后验分布的最佳参数,从而得到关联向量机.建立起基于参数分布多维时间序列预测模型,将每一步迭代过程中的支持向量机输入作为随机变量,考虑数据不确定性的传递,递推得到贝叶斯支持向量机预测输出.由于贝叶斯支持向量机可以有效反映随机影响及其传递,可以克服数据不确定性和相关性的影响,因此基于贝叶斯支持向量机预测效果更加符合实际.实例表明利用贝叶斯支持向量机预测高科技企业发展趋势与实际发展趋势接近,可以克服数据相关性、不确定性和非线性对信息模型的影响,具有较高的预测精度和预测鲁棒性.     

基于SMO—SVR的飞机舵面损伤故障趋势预测

飞机舵面出现损伤时,为了更准确的预测状态参量变化情况,提出了一种改进的序贯最小优化支持向量回归(SMO-SVR, Sequential Minimal Optimization Support Vector Regression)预测方法.采用改进C-C平均方法对多元时间序列进行相空间重构,以确定最优嵌入维数m和延迟时间τ_d.根据所求m和τ_d建立加权SVR预测模型,并调整了SMO算法的停机准则.利用区间自适应粒子群算法(IAPSO, Interval Adaptive Particle Swarm Optimization)优化SVR参数,以提高参数优化速度.为了验证改进算法的有效性,针对飞机方向舵损伤故障趋势进行了预测和分析,并与径向基函数神经网络(RBFNN, Radial Basis Function Neural Network)方法进行了对比,仿真结果表明SMO-SVR预测模型具有很好的预测能力.     

一种改进的近似支持向量机算法

标准近似支持向量机受类别差异影响和噪声、野值数据干扰较重,使得分类能力不高.提出一种改进的近似支持向量机算法——加权近似支持向量机,通过为不同类别设定不同的惩罚参数和为每个样本引入模糊隶属值,有效补偿类别差异带来的倾向性并去除噪声和野值数据的影响.模糊隶属函数的选取采用样本与类中心的距离和样本紧密度的加权平均值计算,以有效去除噪声和野值数据的干扰.经过分析,改进后的算法可近似归结为一种岭回归模型.实验表明,与标准近似支持向量机相比,该算法有更好的分类能力.      

一种基于支持向量机的小波图像压缩方法

提出了一种结合支持向量机(SVM,Support Vector Machines)回归与小波变换的新的静态图像压缩方法.SVM回归方法可以学习原始数据之间的相关性,并采用小部分训练样本,即支持向量来稀疏表示原始数据集,利用这一特性来逼近和约减小波系数,可以达到数据压缩的效果.首先采用小波变换把原始图像分解成不同尺度的多个子带,由于最低频子带系数非常重要,采用DPCM直接编码,然后对其它频带系数采用SVM回归进行压缩.由于不同尺度和方向的小波系数特征不同,为尽可能去除小波系数间的各种相关性,给出了适合SVM回归的小波系数的有效组织方式.最后研究了支持向量及其相应权重的混合编码方法.实验结果表明:与同类压缩方法相比,本算法获得的恢复图像的主客观质量有明显提高.      

基于支持向量机的机场智能驱鸟决策

为提高机场鸟击防范管理水平,实现探鸟雷达与多种驱鸟设备联动,提出一种基于支持向量机（SVM）的机场智能驱鸟决策方法。该方法包括训练和测试两部分。训练部分利用机场鸟类探测预警与驱赶联动系统获取的大量历史鸟情信息,结合专家知识,通过数据预处理与支持向量机训练,建立驱鸟策略分类模型;测试部分根据驱鸟实时智能决策结果,对驱鸟策略分类模型进行持续修正与优化。通过某机场的实测鸟情信息数据与若干驱鸟实例,证明驱鸟策略分类模型具有较高的决策正确率,并能够通过自身修正与优化应对各种新问题。本文方法针对实时鸟情信息,实现了多种驱鸟设备的优化组合,克服了驱鸟设备长期重复运行造成的鸟类对驱鸟设备的耐受性问题,极大改善了驱鸟效果。      

免疫支持向量机用于航空发动机磨损故障诊断

航空发动机在使用寿命周期内会不断磨损最终出现故障,通过对发动机油液监测铁谱分析数据的挖掘可实现磨损故障的诊断。本文研究免疫算法优化的支持向量机(SVM)在航空发动机磨损故障诊断中的运用。首先,总结了支持向量机和免疫算法的运行流程和关键算法。然后,用改进的免疫算法优化支持向量机惩罚因子、松弛变量及核函数参数。某型航空发动机的油液铁谱分析数据和加入噪声数据验证结果表明,该方法可有效实现航空发动机磨损故障诊断且具有较好的鲁棒性。最后,研究了核函数、多分类决策方法、初始种群大小、亲和力计算公式、支持向量机优化方法和归一化方法对磨损故障诊断准确率的影响,得到了最佳诊断方法。     

基于最小偏差LS-SVR的模电系统多故障诊断

为解决支持向量分类机多分类存在的样本重复训练、训练模型过多的问题,保证模拟电子系统在整体和局部多故障模式上的诊断正确率,提出基于最小偏差的最小二乘支持向量回归机多故障诊断方法.通过引进样本各维度拟合误差相对于平均拟合误差的偏差平方项,最小化维度间的拟合误差间距,得到能够输出多维变量及具有高分辨率的最小二乘支持向量回归机模型.将模型多维输出值与预设的各个多故障模式值相乘,所得结果集中最大值所对应的多故障模式即为最终诊断结果.仿真结果表明:提出的方法在简化训练过程的同时,能够保持良好的整体和局部多故障诊断正确率.     

Equatorial F-region vertical ion drifts during quiet solar maximum

Median values of ionosonde h′F data acquired at Ibadan (Geographic:7.4°N, 3.9°E, Magnetic: dip 6°S, and magnetic declination, 3°W), Nigeria, West Africa, have been used to determine vertical ion drift (electric field) characteristics in the postsunset ionosphere in the African region during a time of high solar activity (average F_10.7 −208). The database spans from January and December 1958 during the era of International Geophysical Year (IGY) for geomagnetic quiet conditions. Bimonthly averaged diurnal variations patterns are very similar, but differ significantly in magnitude and in the evening reversal times. Also, monthly variations of F-region vertical ion drift reversal times inferred from the time of h′F maximum indicates early reversal during equinoxes and December solstice months except for the month of April. Late reversal is observed during the June solstice months. The equatorial evening prereversal enhancement in vertical ion drift (V_zp) occurs largely near 1900 LT with typical values 20–45 m/s. Comparison of Ibadan ionosonde V_zp with the values of prereversal peak velocity reported for Jicamarca (South America), Kodaikanal (India), and Scherliess and Fejer global model show considerable disparity. The changes of postsunset peak in virtual height of F-layer (h′F_P) with prereversal velocity peak V_zp are anti-correlated. Investigation of solar effects on monthly values of V_zp and h′F_P revealed that these parameters are independent of monthly averaged solar flux intensity during quiet-time sunspot maximum conditions.     

基于Tophat变换和文字纹理的车牌定位算法

针对车牌自动识别系统的应用场景越来越广泛,提出一种新的基于Tophat滤波和文字纹理特征的车牌定位算法.首先,利用形态学Tophat滤波抑制背景、消除不均匀光照,然后对图像进行二值化操作和连通域分析;其次,扫描图像得到区域的垂直投影图(VPM, Vertical Projection Map),对VPM进行离散余弦变换(DCT, Discrete Cosine Transformation),利用中低频描绘子重构VPM,重构后的VPM更为平滑,不受噪声影响,其波峰波谷数量、统计量更能描绘区域纹理的本质属性;最后,结合部分中低频描绘子和统计量组成描述区域纹理的模式向量,输入支持向量机归类.实验表明,算法适用于自然场景中的车牌定位问题,具有较强的适应性.      

基于SVM的浮动车行驶模式判断模型

浮动车在低速情况下存在两种行驶模式,如不能对上述模式进行准确区分,将严重影响浮动车实时路况计算的精度和效率.研究和设计了一个基于支持向量机(SVM,Support Vector Machine)的浮动车行驶模式判断模型,并针对性地提出了一种简单的基于隶属度矩阵的特征评价和选择方法.实验表明通过上述方法选择的特征子集所训练的分类器在测试样本集上具有92.6%的分类准确性;经过行驶模式分析后,浮动车系统的准确性有显著提升.      

Using the radial basis function neural network to predict ionospheric critical frequency of F2 layer over Wuhan

Neural networks (NNs) have been applied to ionospheric predictions recently. This paper uses radial basis function neural network (RBF-NN) to forecast hourly values of the ionospheric F₂ layer critical frequency(f_oF₂), over Wuhan (30.5N, 114.3E), China. The false nearest neighbor method is used to determine the embedding dimension, and the principal component analysis (PCA) is used to reduce noise and dimension. The whole study is based on a sample of about 26,000 observations of f_oF₂ with 1-h time resolution, derived during the period from January 1981 to December 1983. The performance of RBF-NN is estimated by calculating the normalized root-mean-squared (NRMSE) error, and its results show that short-term predictions of f_oF₂ are improved.     

一种基于SVM的低空飞行冲突探测算法

随着低空飞行密度不断增加,低空航行安全已引起广泛关注,由于低空环境复杂,低空飞行受地面障碍物和天气影响比商用航空显著,传统的空中交通警戒与防撞系统（TCAS）和其他冲突探测方法并不适用于低空密集飞行环境。针对传统探测方法计算量大、适用性差的不足,引入支持向量机（SVM）的二元分类方法,通过对本机和周边飞机航迹归一化处理,采用智能优化算法对关键参数进行优化,利用模拟数据对分类器进行预先训练,实现了适用于低空飞行的高效冲突探测。以大量的仿造数据对算法有效性进行了测试验证,结果表明漏警率和误警率分别控制在约0.1%和6%,克服了传统确定型方法与概率型方法难以兼顾效率与适用性的缺陷。      

Comparison of observed hmF2 and IRI 2007 model with M(3000)F2 estimation of hmF2 at low solar activity for an equatorial station

This is to investigate ways of improving the Equatorial F2-layer peak heights estimated from M(3000)F2 ionosonde data measured using the Ionospheric Prediction Service (IPS-42) sounder at Ouagadougou, Burkina Faso (Latitude +12.4°N, Longitude +1.5°W, Dip latitude +5.9°N) during a low solar activity year (1995). For this purpose, we have compared the observed h_mF2 (h_mF2_obs) deduced using an algorithm from scaled virtual heights of quiet day ionograms and the predicted h_mF2 values which is given by the IRI 2007 model (h_mF2_{IRI 2007}) with the ionosonde measured M(3000)F2 estimation of the h_mF2 values (h_mF2_est) respectively. The correlation coefficients R² for all the seasons were found to range from 0.259 to 0.692 for h_mF2_obs values, while it ranges from 0.551 to 0.875 for the h_mF2_{IRI 2007} values. During the nighttime, estimated h_mF2 (h_mF2_est) was found to be positively correlated with the h_mF2_obs values by the post-sunset peak representation which is also represented by the h_mF2_{IRI 2007} values. We also investigated the validity of the h_mF2_est values by finding the percentage deviations when compared with the h_mF2_obs and h_mF2_{IRI 2007}.     

A different method to update monthly median hmF2 values

The height, h_mF2, and the electron density, N_mF2, of the F2 peak are key model parameters to characterize the actual state of the ionosphere. These parameters, or alternatively the propagation factor, M3000F2, and the critical frequency, f_oF2, of the F2 peak, which are related to h_mF2 and N_mF2, are used to anchor the electron density vertical profile computed with different models such as the International Reference Ionosphere ( Bilitza, 2002), as well as for radio propagation forecast purposes. Long time series of these parameters only exist in an inhomogeneous distribution of points over the surface of Earth, where dedicated instruments (typically ionosondes) have been working for many years. A commonly used procedure for representing median values of the aforementioned parameters all over the globe is the one recommended by the ITU-R ( ITU-R, 1997). This procedure, known as the Jones and Gallet mapping technique, was based on ionosondes measurements gathered from 1954 to 1958 by a global network of around 150 ionospheric stations (  and ). Even though several decades have passed since the development of that innovative work, only few efforts have been dedicated to establish a new mapping technique for computing h_mF2 and N_mF2 median values at global scale or to improve the old method using the increased observational database. Therefore, in this work three different procedures to describe the daily and global behavior of the height of the F2 peak are presented. All of them represent a different and simplified method to estimate h_mF2 and are based on different mathematical expressions. The advantages and disadvantages of these three techniques are analyzed, leading to the conclusion that the recommended procedure to represent h_mF2 is best characterized by a Spherical Harmonics expansion of degree and order equal to 15, since the differences between the h_mF2 values obtained with the Jones and Gallet technique and those obtained using the abovementioned procedure are of only 1%.     

An alternative Shell inversion technique - Analysis and validation based on COSMIC and ionosonde data

Multi-channel Global Positioning System (GPS) carrier phase signals, received by the six low Earth orbiting (LEO) satellites from the Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) program, were used to undertake active limb sounding of the Earth’s atmosphere and ionosphere via radio occultation. In the ionospheric radio occultation (IRO) data processing, the standard Shell inversion technique (SIT), transformed from the traditional Abel inversion technique (AIT), is widely used, and can retrieve good electron density profiles. In this paper, an alternative SIT method is proposed. The comparison between different inversion techniques will be discussed, taking advantage of the availability of COSMIC datasets. Moreover, the occultation results obtained from the SIT and alternative SIT at 500 km and 800 km, are compared with ionosonde measurements. The electron densities from the alternative SIT show excellent consistency to those from the SIT, with strong correlations over 0.996 and 0.999 at altitudes of 500 km and 800 km, respectively, and the peak electron densities (N_mF₂) from the alternative SIT are equivalent to the SIT, with 0.839 vs. 0.844, and 0.907 vs. 0.909 correlation coefficients when comparing to those by the ionosondes. These results show that: (1) the N_mF₂ and h_mF₂ retrieved from the SIT and alternative SIT are highly consistent, and in a good agreement with those measured by ionosondes, (2) no matter which inversion technique is used, the occultation results at the higher orbits (∼800 km) are better than those at the lower orbits (∼500 km).