随机森林结合直接正交信号校正的模型传递方法

为增强近红外光谱模型通用性,解决直接正交信号校正算法在光谱处理过程中可能出现过拟合、模型不稳定的现象,提出一种将随机森林与直接正交信号校正算法相结合的模型传递方法(Random Forest-Direct Orthogonal Signal Correction,RF-DOSC)。该方法首先利用随机森林算法进行近红外光谱波长点筛选,然后采用直接正交信号校正方法进行光谱处理并建立回归方程,由PLS计算回归系数求得模型传递矩阵。实验使用三台光谱仪(S,S1,S2)测得的玉米近红外光谱数据集建立传递模型,数据集1(D1)水分、油分、蛋白质、淀粉成分预测标准偏差(SEP)分别为0.1267、0.0982、0.1569和0.4051,数据集2(D2)四种成分的SEP分别为0.1548、0.0819、0.1366和0.3836,均小于传统方法。实验结果表明本文所提模型传递方法能有效消除光谱噪声,减小主仪器和从仪器光谱之间的差异,提高模型的稳定性和准确性,实现不同仪器之间模型的共享。     

基于蒙特卡罗无信息变量消除的烟气指标预测

使用近红外光谱(NIRS)分析方法对烟叶的CO、烟碱、焦油含量进行无损快速定量分析,可以提高分析方法的预测精度,消除无信息建模变量对模型稳定性的影响。在此以烟叶为研究对象,利用蒙特卡罗无信息变量消除方法(MC-UVE)对烟叶的近红外光谱进行了波段点的筛选,并利用筛选出的波段建立PLS校正模型。结果表明利用蒙特卡罗无信息变量消除方法可以有效选择建模变量,既克服了复杂样品各信息区间对PLS建模贡献率不一样的问题,又能提高模型的稳定性和多元校正的预测精度。     

广东省主要人工林树种光谱分类

为了探索不同树种叶片光谱区分的最佳特征波段以及不同树种光谱分类的 性能,利用地物光谱仪对广东省10个主要人工林树种的叶片光谱数进行了采集。采用遗 传算法(Genetic Algorithm, GA)和连续投影变换算法(Successive Projections Algorithm, SPA) 进行了高光谱数据降维处理,然后结合支持向量机(Support Vector Machine, SVM)和随机森林(Random Forest, RF) 两种方法进行了树种分类。研究发现,通过两种变量选择方式筛选的用于树种分类的光谱范围 主要位于近红外波段。其中,经GA算法筛选的变量建模和预测精度与基于全波段光谱数据的分类精 度较为接近,且比经SPA算法变量筛选的分类结果好。通过比较可知,RF算法所建模型的性能比SVM算 法更稳定,且GA-RF算法在几种分类处理中的性能最好。结果表明,GA-RF算法 可用于基于光谱数据的树种分类研究。     

基于改进模糊均值聚类算法的医学图像分割

为了提高医学图像分割性能,针对传统模糊聚类算法存在的缺陷,提出了一种改进模糊均值聚类算法的医学图像分割方法。首先采用粒子群算法选择模糊均值聚类算法的聚类中心,然后利用空间邻域信息设定聚类样本空间,最后采用具体的医学图像数据进行仿真实验,测试其有效性。仿真结果表明,相对于传统模糊聚类算法,本文算法不仅提高了医学图像分割精度,而且提高了医学图像分割效率。     

基于可见-近红外光谱特征波长选择的土壤有机质快速检测研究

选择光谱特征波长进行建模可以减少冗余波长的干扰,提高模型的预测精度。采用小波阈值消噪法对采集的104个土壤样本光谱数据进行了预处理,并通过间隔偏最小二乘法、无信息变量消除、连续投影算法和群智能算法等9种方法筛选了建模波长。结果表明,小波阈值消噪法能有效降低光谱中的噪声。利用波长选择方法筛选建模波长不仅能减少建模变量的个数,而且还能提高模型的预测精度,特别是离散粒子群优化算法利用26个波长进行建模,预测决定系数达到了0.81,预测的相对标准误差为2.31。实验结果证明,通过对光谱波长进行选择不但可以降低模型的复杂度,还能有效预测土壤有机质达的含量。     

RX及其变种在高光谱图像中的异常检测

为了提高核RX算法在高光谱图像异常检测中的稳定性,将核矩阵正则化,并提出正则化的核RX算法(rkRX)。将规范化后的正则化RX算法和正则化的核RX算法融合改进,称为融合RX算法(mRX),该算法同时考虑了原始线性空间和高维特征空间的异常检测结果,使异常检测效果更加稳定。在仿真图像和真实高光谱图像的实验中,上述2种算法与原始的RX、正则化RX(rRX)和核RX(kRX)3种算法进行了比较,使用了双窗口技术和核主成分分析(KPCA)进行特征提取和基于高阶统计量的特征选择作为预处理来降低数据维数,并在未降维数据上比较上述5种算法。最后,使用ROC曲线评价检测效果,结果表明:提出的2种算法提高了检测效果并具有一定鲁棒性。     

光谱配置对最优波段组合算法预测土壤有机质和电导率的影响

可见光和近红外(Vis-NIR)光谱仪是一种实用的工具,可用于替代土壤物理和化学分析以评估土壤性质.最佳波段组合算法是一种通过考虑波段之间的相互作用信息来提取光谱变量的有效方法,但是对于实验室Vis-NIR光谱数据,该方法易受"维数灾难"的影响.提出一种适当地降低光谱配置(减少光谱带的数量和粗化光谱分辨率),以提高计算效率且不会显著影响预测精度的方法.首先,构建了6个光谱配置,即光谱波段的数量从2001个减小到19个,光谱分辨率从3 nm增加至100 nm,并且光谱采样间隔等于光谱分辨率(均匀间隔采样).然后,通过偏最小二乘回归结合最佳波段组合算法预测不同光谱配置下土壤有机质(SOM)和电导率(EC)的最佳光谱参数.结果 表明:直到光谱分辨率为60 nm时(每个光谱32个波长),最佳波段组合算法与全波段光谱数据相比,可以提高预测精度.最佳波段组合算法在1～20 nm光谱分辨率下,预测精度没有显著差异(约为2％);SOM[(决定系数R2v等于0.77,均方根误差(RMSEP)等于0.325％,性能四分位数范围的比率(RPIQv)等于3]和EC(R2v等于0.70,RMSEP等于6.88 dS·m-1,RPIQv等于2.21)分别在20 nm和10 nm的光谱分辨率下获得了最佳预测性能.     

基于IRIV-SA的乙烯FTIR光谱波数优选

利用傅里叶变换红外光谱(FTIR)技术能够对有机物的成分及含量进行快速无损测量。作为重要的有机化工原料,乙烯在塑料、醇类和纤维等大宗化学品的制造中有着广泛的应用,但同时由于其易挥发性,乙烯对环境和人体有着潜在的危害。为提高FTIR技术检测乙烯浓度模型的精度,综合迭代保留信息变量法(IRIV)和模拟退火算法(SA)的优点,提出了改进的IRIV-SA红外光谱波数优选算法。该方法在IRIV算法稳定选取大量光谱特征波数的基础上,利用SA进一步筛选少量有效特征波数,从而降低模型复杂度,提高有机物光谱检测精度。实验首先利用IRIVSA对乙烯红外光谱的浓度进行波数选取,最终获取的特征波数由全光谱的271个变量降低至5个变量,再利用特征波数进行建模,结果表明其验证集相关系数、均方根误差为0.9989和0.3943,预测集相关系数、均方根误差为0.9978和0.6652,较全光谱建模精度有大幅提高。为进一步验证该算法的有效性,同时建立IRIV、SA、CARS (自适应重加权采样算法)、SPA (连续投影算法)以及IRIV-CARS、IRIV-SPA波数选取模型对相同数据集进行对比实验,比对结果表明IRI...     

基于可见-近红外光谱技术与BP-ANN算法的污水类型鉴定

提出了一种基于可见-近红外光谱技术与BP人工神经网络(BP-ANN)算法快速进行污水类型鉴定的新方法.以FieldSpec(R)3地物光谱仪采集了4种污水样品的光谱数据,共168份,随机将其分成校正集(132份)和检验集(36份).分别采取全波段(400～2450 nm)与择取波段(400～1800 nm)两种方法建立模型进行分析.光谱经S.Golay平滑和标准归一化(SNV)处理后,以主成分分析法(PCA)降维.将降维所得的前9个主成分数据作为BP-ANN的输入变量,污水类型作为输出变量,建立3层BP-ANN鉴别模型.利用36个未知样对模型进行检验.结果表明:两类模型预测准确率均高达100%,且择取波段模型比全波段模型具有更高的预测精度.说明利用可见-近红外技术结合BP-ANN算法进行污水类型的快速、无污染鉴定是可行的,且波段筛选是优化模型的有效方法之一.     

多光谱图像技术在土壤酸碱度检测中的应用

提出了一种利用多光谱图像的颜色特征对土壤酸碱度(pH值)进行快速无损检测的方法。首先,利用2CCD多光谱成像仪获取每个土壤样本的R、G、B、NIR图像各一幅,并对多光谱图像进行颜色空间转换,即从RGB色彩空间分别转换到HSV颜色空间和Lab颜色空间;然后提取不同颜色空间中多光谱图像的颜色特征;最后,分别将提取的颜色特征作为模型的输入变量,建立PLS和LS-SVM算法的土壤酸碱度预测模型。实验结果表明,利用多光谱图像技术对土壤酸碱度进行检测是可行的。预测模型的最佳结果如下:决定系数(R~2)为0.91,预测均方根误差(RMSEP)为0.34.     

An optimal designed degradation experiment for reliability improvement

In conducting a design-of-experiments, with degradation data, to improve the reliability of highly reliable products, one of the most important topics for manufacturers is to identify important factors that affect the product reliability. Several decision variables such as inspection frequency, sample size, and termination time at each run, do influence the precision of identifying the important factors and the experimental cost. This paper proposes a systematic approach to the identification problem with respect to fractional factorial degradation experiments. First, an intuitively appealing identification rule is proposed. Next, the optimum test plan is derived using the criterion of minimizing the total experimental cost. The inspection frequency, sample size, and termination time at each run, that are needed by this identification rule are computed by solving a nonlinear integer programming model with a minimum probability of correct selection. A simulation study investigates the efficiency of this method.     

基于深层次挖掘算法的台区供电相位分析方法

在台区供电相位分析中,传统的分析方法面对由杂乱数据过多造成的频率密集、含有偶次谐波或间谐波的情况,稳定性极差。为此提出基于深层次挖掘算法的台区供电相位分析方法。利用挖掘算法挖掘出台区用电数据关联规则,清除无用数据,使用数据信号频率合成技术合成信号;利用傅里叶变换获得全相位输入数据,通过相移量补偿相位差,再计算出振幅、相位和频率,实现台区供电相位分析。实验结果表明,在相同的实验条件下,设计的基于深层次挖掘算法的台区供电相位分析方法抗噪声性能强,参数误差小,稳定性高,适合应用在实际项目中。     

基于类别分布差异和VPRS特征选择的文本分类方法

权值计算和特征降维是影响文本分类的精度和效率的两个重要步骤。该文首先根据特征词的类别分布差异进行特征过滤;然后,分析传统的权值公式TF-IDF的缺点,采用改进的权值计算公式简记为TF-CDF,依据TF-CDF公式计算每个特征词的权值,生成文档集的向量空间模型VSM;接着,提出了一种基于可变精度粗糙理论(VPRS)的特征选择进一步选择对分类贡献度大的特征,并用SQL实现。最后利用支持向量机LibSVM分类器进行实验,实验结果表明特征过滤和选择方法及TF-CDF权值公式有助于提高分类精度和分类效率。     

Input variable selection for feature extraction in classification problems

We propose an input variable selection method based on discriminant features. By analyzing the relationship between the input space and feature space obtained by discriminant analysis, the input variables that contain a large amount of discriminative information are selected, while input variables with less discriminative information are discarded. By this, the signal to noise ratio of the data can be improved. The proposed method can be applied not only to the feature extraction methods based on covariance matrix but also to the methods based on image covariance matrix. The experimental results obtained with various data sets show that the proposed method results in improved classification performance regardless of the dimension and type of data.     

自适应陷波滤波器频率估计新方法及性能分析

现有的自适应陷波滤波器（ANF）受误差函数所限,导致其自适应频率估计方法收敛速度较慢,对初始迭代频率值设定范围要求较高,特别针对频率接近于0或π的信号,还存在频率估计精度不高、算法稳定性差的问题,为此,提出一种ANF频率估计新方法.首先,分析现有ANF方法估计信号频率时存在精度低、速度慢、稳定性差的原因,提出一种新误差函数以提升ANF收敛速度;然后,根据ANF估计信号频率时偏差产生的机理,通过偏差补偿方式,降低噪声对ANF的影响,以获得近似无偏的频率估计结果,提高ANF频率估计精度,同时与离散卡尔曼滤波相结合,以改善算法的稳定性,并对该方法进行稳态条件下的性能分析;最后,给出了ANF频率估计结果,并讨论了ANF各参数对频率估计精度的影响,给出了具体计算结果.计算表明本文方法的有效性与正确性.     

Network traffic prediction method based on improved ABC algorithm optimized EM-ELM

In order to overcome the poor generalization ability and low accuracy of traditional network traffic prediction methods, a prediction method based on improved artificial bee colony (ABC) algorithm optimized error minimized extreme learning machine (EM-ELM) is proposed. EM-ELM has good generalization ability. But many useless neurons in EM-ELM have little influences on the final network output, and reduce the efficiency of the algorithm. Based on the EM-ELM, an improved ABC algorithm is introduced to optimize the parameters of the hidden layer nodes, decrease the number of useless neurons. Network complexity is reduced. The efficiency of the algorithm is improved. The stability and convergence property of the proposed prediction method are proved. The proposed prediction method is used in the prediction of network traffic. In the simulation, the actual collected network traffic is used as the research object. Compared with other prediction methods, the simulation results show that the proposed prediction method reduces the training time of the prediction model, decreases the number of hidden layer nodes. The proposed prediction method has higher prediction accuracy and reliable performance. At the same time, the performance indicators are improved.     

基于PLS成分的变量筛选法

就多元线性回归模型，提出一种基于PLS成分的变量筛选法。其基本思想是利用偏最小二乘回归原理，提取系统的两个主要成分，然后根据变量入选原则，保留对因变量影响较大的变量；再利用PLS成分筛选剩下的变量，经有限次筛选，即可得到最佳变量子集。应用算例证实了这一方法的有效性和可行性。     

Semiconductor Thermoelectric Converters of Alternating Current Signals

Measuring converters have been developed based on Bi₂Te₃ solid solutions optimized for temperature and temporal stability of the basic thermoelectric parameters (thermal conductivity, electrical conductivity, and the Seebeck coefficient). Semiconductor material was grown by the vertical zone melting method. High efficiency of the optimized semiconductor materials made it possible to reduce the operating temperature considerably, as well as facilitating selection of structural materials and the converter fabrication technique. The influence of the Peltier and Thomson effects on the accuracy of direct current conversion has been reduced, and the operating frequency range of measurements has been expanded. Thermoelectric converters with sensitivity up to 30 V W⁻¹ have been created without evacuation of the working space, which is filled with an inert gas mixture. The engineered converters offer the advantages of high sensitivity and a wide operating frequency range (up to 30 MHz).     

Bagging linear sparse Bayesian learning models for variable selection in cancer diagnosis.

This paper investigates variable selection (VS) and classification for biomedical datasets with a small sample size and a very high input dimension. The sequential sparse Bayesian learning methods with linear bases are used as the basic VS algorithm. Selected variables are fed to the kernel-based probabilistic classifiers: Bayesian least squares support vector machines (BayLS-SVMs) and relevance vector machines (RVMs). We employ the bagging techniques for both VS and model building in order to improve the reliability of the selected variables and the predictive performance. This modeling strategy is applied to real-life medical classification problems, including two binary cancer diagnosis problems based on microarray data and a brain tumor multiclass classification problem using spectra acquired via magnetic resonance spectroscopy. The work is experimentally compared to other VS methods. It is shown that the use of bagging can improve the reliability and stability of both VS and model prediction.     

Joint beam selection and user association strategy for beam space millimetre wave downlink systems

Beam space multiple‐input multiple‐output (B‐MIMO) systems operating at mmWave frequency in downlink uses different beam selection techniques for reducing the required number of radio frequency chains without apparent performance loss. In this paper, a joint beam selection and user association scheme with the objective of sum rate maximization is proposed. The proposed work on beam selection is based on a norm and uncorrrelation based approach, which considers channel correlation among the users so as to avoid inter‐user interference. Two different methods for beam user association with one modelled as a balanced assignment problem based on average sum assignment method and the other method that associates an optimal beam to every user by considering their channel gain values are also proposed along with beam selection. The proposed beam assignment methods are less complex and efficient in finding a suitable beam to every user within the cell. Simulation results show that the proposed method accomplish comparable performance in terms of sum rate and energy efficiency with appreciably low computational complexity relative to the existing methods in a correlated environment.