面向不平衡数据集的一种基于聚类的欠采样方法

针对不平衡数据集分类问题,提出了一种基于聚类的欠采样方法.分别取不同的聚类个数,对训练集中的多数类样本进行若干次聚类,然后用聚类中心作为多数类样本,与少数类样本构成若干个新的训练集,之后用这些训练集训练分类器,剔除具有错误分类倾向的分类器,最后对分类结果进行投票.仿真实验对几种欠采样方法进行比较.实验采用16个平衡率不一的数据集进行测试.理论分析与实验结果表明:提出的基于聚类的欠采样方法能有效地改善不平衡数据集的不平衡性.     

基于机器学习的黏钢构件黏接层缺陷识别方法研究

对黏钢加固结构黏接层缺陷对超声检测信号的影响进行了深入研究,并提出了一种基于机器学习的黏接层缺陷识别的新型方法.首先,该文基于直接接触式的脉冲回波反射法对黏钢构件进行有限元模拟,并阐述了超声波在黏钢构件中的传播规律;其次,通过分析局部段超声回波信号及相关信号特征,讨论了不同缺陷变量对超声回波信号的影响规律;最后,建立了黏钢构件超声时程响应数据集,并对比了不同机器学习模型对缺陷大小、位置的分类识别性能,形成了黏钢构件黏接层缺陷识别方法.结果表明,局部段超声回波信号及其特征随着缺陷大小、位置的改变呈规律性变化,能够对缺陷信息进行初步区分.同时,该文提出的基于RF模型的黏钢构件黏接层缺陷识别方法能够有效识别黏钢构件黏接层缺陷,具有较广阔的工程应用前景.     

RBF神经网络电流保护

传统继电器电流保护方案在实施的时候,受到的影响要素主要是来源于电力系统采用何种方式运行,以及各种常见的故障类型,而自适应电流保护解决方案可以很好的避免传统保护出现的弊端和缺陷.在此基础之上构建出一种基于RBF神经网络电流保护解决方案,这种系统主要运用的网络包含了3层RBF神经网络模型,主要的部分构成拥有三种不同的类型:故障方向判别、故障类型与相别判断.为了能够确保该系统能够可靠的实现保护,对其进行了深度的测试和各类型的仿真实验,通过最终获取的实验数据,分析之后得到RBF网络训练效率高效同时证明该电流保护方案在实践当中具备可行性.     

基于自步学习与稀疏自表达的半监督分类方法

基于图的半监督分类方法近年来在模式识别和机器学习领域取得了广泛的关注.然而许多传统方法在构建邻域图时采用固定的邻域尺寸,且在模型训练过程中同等对待所有样本,忽略了样本间的差异性,从而影响了方法的效果.对此,文章提出一种基于自步学习和稀疏自表达的半监督分类方法,提取并保持数据的有判别信息的稀疏自表达结构,并基于自步学习机制提出一种新的自步学习项,将数据重要程度的软权重与硬权重相结合,来对样本进行学习.所提方法能够自适应建立数据间的关系,自动给出样本的重要程度并由易到难进行学习,且具有多类的显性非线性分类函数.几个标准数据集上的实验结果表明,所提算法具有较好的半监督分类效果.     

高斯型模糊前向神经网络

利用高斯型隶属函数和采样数据得到了三层模糊前向神经网络。该网络模型利用权值直接确定法得到了最优权值,并依据采样数据中的插值样本较好确定了单隐层神经元个数。该网络是近似插值神经网络。仿真实验表明,高斯型模糊前向神经网络具有逼近精度高、网络结构简单、良好的去噪性和实时性高等优点。     

不平衡数据处理的新方法——基于样本相似度的少数类合成法

不平衡数据是指分类问题中目标变量的某一类观测值数量远大于其他类观测值数量的数据。针对处理不平衡数据算法SMOTE及其衍生算法的不足,本文提出一种新的向上采样算法SMUP(Synthetic Minority Using Proximity of Random Forests),通过样本相似度改进SMOTE算法中的距离测量方式,提高了算法的分类精度。实验结果表明,基于SMUP算法的单分类器能有效提升少数类的分类正确率,同时解决了SMOTE对定类型特征变量距离测度不佳的难题;基于SMUP算法的组合分类器分类效果也明显优于SMOTE衍生算法;最重要的是,SMUP将连续型、混合型和定类型这三种特征变量的距离测度整合到一个统一的框架下,为实际应用提供了便利。     

基于FPGA的自适应TIADC数据采集

通过对并行时间交替采样(Time-interleaved ADC,TIADC)理论和采样时间失配误差的研究,提出一种带时间补偿的自适应TIADC数据采集方案.方案首先产生具有等相位差的采样时钟,通过自适应电路估计时间失配误差;然后利用Farrow结构的分数延时滤波器对多通道采集数据进行校正,并交叉合并,得到高速TIADC数据;最后通过FPGA实验验证,证明自适应TIADC数据采集不仅可以提高采样率,而且能有效校正时间失配误差,保证数据采集的准确性.     

应用模糊支持向量机进行英文情感分类

针对英文情感分类问题,对不同样本采用不同权重,通过引入模糊隶属度函数,通过计算样本模糊隶属度确定样本隶属某一类程度的模糊支持向量机分类算法,通过对比选取不同核函数和不同惩罚系数的结果.仿真实验结果表明应用模糊支持向量机进行英文情感分类具有较好的分类能力和较高的识别能力.     

基于改进Adaboost的信用评价方法

网络借贷环境下基于Adaboost的信用评价方法具有较高的基分类器分歧度和样本误分代价。现有研究没有考虑分歧度和误分代价对基分类器样本权重的影响,从而降低了网络借贷信用评价结果的有效性。为此,提出一种基于改进Adaboost的信用评价方法。该方法根据基分类器的误分率,样本在不同基分类器上分类结果的分歧程度,以及样本的误分代价等因素,调整Adaboost模型的样本赋权策略,使得改进后的Adaboost模型能够对分类困难样本和误分代价高的样本实施有针对性的学习,从而提高网络借贷信用评价结果的有效性。基于拍拍贷平台数据的实验结果表明,提出的方法在分类精度和误分代价等方面显著优于传统的基于Adaboost的信用评价方法。     

基于准ARX多层学习网络模型的非线性系统自适应控制

建立了准ARX多层学习网络预测模型,并用于非线性系统自适应控制问题.该模型的内核部分为一个改进的神经模糊网络(NFNs):一部分为三层非线性网络结构,采用自联想网络进行离线训练;另一部分为三层NFNs,采取在线调整.据此对参数进行分类,给出相应调整算法. 然后,基于模型宏观结构的优势给出控制器设计方案.仿真分析给出该建模方法的有效性.     

考虑多种类型缺陷的软件可靠性模型研究

本文提出一种考虑多种类型缺陷的软件可靠性模型,并构建了缺陷检测和剔除两个过程的模型。具体分类情况,可以根据模型的检验方法(拟合准则和预测有效性度量)和模型复杂度来具体决定,如果有测试人员的分类建议或者分类数据,可以结合模型共同决定。为了说明问题,本文给出四种类型缺陷的具体模型,并对实际数据集进行了拟合。通过模型比较,验证了多种类型缺陷模型的有效性。最后,通过构建软件最优发布时间策略对模型进行了应用。研究结果为软件开发和测试提供了理论参考。     

Classification of underwater signals using wavelet transforms and neural networks

Neural network classifiers have been widely used in classification due to its adaptive and parallel processing ability. This paper concerns classification of underwater passive sonar signals radiated by ships using neural networks. Classification process can be divided into two stages: one is the signal preprocessing and feature extraction, the other is the recognition process. In the preprocessing and feature extraction stage, the wavelet transform (WT) is used to extract tonal features from the average power spectral density (APSD) of the input data. In the classification stage, two kinds of neural network classifiers are used to evaluate the classification results, inclusive of the hyperplane-based classifier—Multilayer Perceptron (MLP)—and the kernel-based classifier—Adaptive Kernel Classifier (AKC). The experimental results obtained from MLP with different configurations and algorithms show that the bipolar continuous function possesses a wider range and a higher value of the learning rate than the unipolar continuous function. Besides, AKC with fixed radius (modified AKC) sometimes gives better performance than AKC, but the former takes more training time in selecting the width of the receptive field. More important, networks trained with tonal features extracted by WT has 96% or 94% correction rate, but the training with original APSDs only have 80% correction rate.     

人工神经网络BP算法的改进和结构的自调整

本文解决了BP神经网络结构参数和学习速率的选取问题,并对传统的BP算法进行了改进,提出了BP神经网络动态全参数自调整学习算法,又将其编制成计算机程序,使得隐层节点和学习速率的选取全部动态实现,减少了人为因素的干预,改善了学习速率和网络的适应能力。计算结果表明：BP神经网络动态全参数自调整算法较传统的方法优越。训练后的神经网络模型不仅能准确地拟合训练值,而且能较精确地预测未来趋势。     

An adaptive neuro‐fuzzy inference system for islanding detection in wind turbine as distributed generation

This article proposes a new integrated diagnostic system for islanding detection by means of a neuro‐fuzzy approach. Islanding detection and prevention is a mandatory requirement for grid‐connected distributed generation (DG) systems. Several methods based on passive and active detection scheme have been proposed. Although passive schemes have a large non‐detection zone (NDZ), concern has been raised on active method due to its degrading power‐quality effect. Reliably detecting this condition is regarded by many as an ongoing challenge as existing methods are not entirely satisfactory. The main emphasis of the proposed scheme is to reduce the NDZ to as close as possible and to keep the output power quality unchanged. In addition, this technique can also overcome the problem of setting the detection thresholds inherent in the existing techniques. In this study, we propose to use a hybrid intelligent system called ANFIS (the adaptive neuro‐fuzzy inference system) for islanding detection. This approach utilizes rate of change of frequency (ROCOF) at the target DG location and used as the input sets for a neuro‐fuzzy inference system for intelligent islanding detection. This approach utilizes the ANFIS as a machine learning technology and fuzzy clustering for processing and analyzing the large data sets provided from network simulations using MATLAB software. To validate the feasibility of this approach, the method has been validated through several conditions and different loading, switching operation, and network conditions. The proposed algorithm is compared with the widely used ROCOF relays and found working effectively in the situations where ROCOF fails. Simulation studies showed that the ANFIS‐based algorithm detects islanding situation accurate than other islanding detection algorithms. © 2014 Wiley Periodicals, Inc. Complexity 21: 10–20, 2015     

Learning Bayesian network classifiers by risk minimization

Bayesian networks (BNs) provide a powerful graphical model for encoding the probabilistic relationships among a set of variables, and hence can naturally be used for classification. However, Bayesian network classifiers (BNCs) learned in the common way using likelihood scores usually tend to achieve only mediocre classification accuracy because these scores are less specific to classification, but rather suit a general inference problem. We propose risk minimization by cross validation (RMCV) using the 0/1 loss function, which is a classification-oriented score for unrestricted BNCs. RMCV is an extension of classification-oriented scores commonly used in learning restricted BNCs and non-BN classifiers. Using small real and synthetic problems, allowing for learning all possible graphs, we empirically demonstrate RMCV superiority to marginal and class-conditional likelihood-based scores with respect to classification accuracy. Experiments using twenty-two real-world datasets show that BNCs learned using an RMCV-based algorithm significantly outperform the naive Bayesian classifier (NBC), tree augmented NBC (TAN), and other BNCs learned using marginal or conditional likelihood scores and are on par with non-BN state of the art classifiers, such as support vector machine, neural network, and classification tree. These experiments also show that an optimized version of RMCV is faster than all unrestricted BNCs and comparable with the neural network with respect to run-time. The main conclusion from our experiments is that unrestricted BNCs, when learned properly, can be a good alternative to restricted BNCs and traditional machine-learning classifiers with respect to both accuracy and efficiency.     

Identification and control of nonlinear systems by a dissimilation particle swarm optimization-based Elman neural network

In this paper, we first present a learning algorithm for dynamic recurrent Elman neural networks based on a dissimilation particle swarm optimization. The proposed algorithm computes concurrently both the evolution of network structure, weights, initial inputs of the context units, and self-feedback coefficient of the modified Elman network. Thereafter, we introduce and discuss a novel control method based on the proposed algorithm. More specifically, a dynamic identifier is constructed to perform speed identification and a controller is designed to perform speed control for Ultrasonic Motors (USM). Numerical experiments show that the novel identifier and controller based on the proposed algorithm can both achieve higher convergence precision and speed than other state-of-the-art algorithms. In particular, our experiments show that the identifier can approximate the USM's nonlinear input–output mapping accurately. The effectiveness of the controller is verified using different kinds of speeds of constant, step, and sinusoidal types. Besides, a preliminary examination on a randomly perturbation also shows the robust characteristics of the two proposed models.     

多层感知器信用评模型及预警研究

本文利用多层感知器 ( MLP)原理建立神经网络信用评价模型 ,用来对我国 2 0 0 0年 1 0 6家上市公司进行信用评级 ,并进一步对我国 2 0 0 1年公布的 1 3家预亏公司进行预警研究 .按照各上市公司的经营状况分为“好”、“差”两类 ,每一类由 5 3家上市公司构成数据样本 .对于每一家上市公司 ,主要考虑其经营状况的四个财务指标 :每股收益 ,每股净资产 ,净资产收益率和每股现金流量 .仿真结果表明 ,本文所建立的神经网络信用评价模型有很高的分类准确率 ,达到 98.1 1 % .又由于该信用评价模型有很强的适应能力 ,故可以进一步用来对企业的财务危机进行预警研究 .预警实证分析表明 ,该信用评价模型对我国 2 0 0 1年公布的 1 3家预亏公司进行预警分析 ,预警准确率达到 1 0 0 % .此外 ,文中还给出 MLP网络模型的学习算法和步骤     

Modeling of Ultrasonic Waves in Fractured Rails with an Explicit Approach

Ultrasonic wave propagation in steel rails with explicit identification of flaws is numerically simulated. The problem is to detect a vertical crack in a railhead by applying ultrasonic nondestructive testing techniques. The propagation of elastic waves in the rail profile is simulated for various sizes and positions of the crack. It is shown that the finite-difference grid-characteristic method in the time domain and full-wave simulation can be used to analyze the effectiveness of rail flaw detection by applying ultrasonic nondestructive testing techniques. Full-wave simulation is also used to demonstrate the failure of the widely used echo-mirror method to detect flaws of certain types. It is shown that techniques for practical application of the ultrasonic delta method can be developed using full-wave supercomputer simulation. The study demonstrates a promising potential of geophysical methods as adapted to the analysis of ultrasonic nondestructive testing results.

     

Fuzzy relational neural network

In this paper a fuzzy neural network based on a fuzzy relational “IF-THEN” reasoning scheme is designed. To define the structure of the model different t-norms and t-conorms are proposed. The fuzzification and the defuzzification phases are then added to the model so that we can consider the model like a controller. A learning algorithm to tune the parameters that is based on a back-propagation algorithm and a recursive pseudoinverse matrix technique is introduced. Different experiments on synthetic and benchmark data are made. Several results using the UCI repository of Machine learning database are showed for classification and approximation tasks. The model is also compared with some other methods known in literature.     

Iterative adaptive RBF methods for detection of edges in two-dimensional functions

Radial basis functions have gained popularity for many applications including numerical solution of partial differential equations, image processing, and machine learning. For these applications it is useful to have an algorithm which detects edges or sharp gradients and is based on the underlying basis functions. In our previous research, we proposed an iterative adaptive multiquadric radial basis function method for the detection of local jump discontinuities in one-dimensional problems. The iterative edge detection method is based on the observation that the absolute values of the expansion coefficients of multiquadric radial basis function approximation grow exponentially in the presence of a local jump discontinuity with fixed shape parameters but grow only linearly with vanishing shape parameters. The different growth rate allows us to accurately detect edges in the radial basis function approximation. In this work, we extend the one-dimensional iterative edge detection method to two-dimensional problems. We consider two approaches: the dimension-by-dimension technique and the global extension approach. In both cases, we use a rescaling method to avoid ill-conditioning of the interpolation matrix. The global extension approach is less efficient than the dimension-by-dimension approach, but is applicable to truly scattered two-dimensional points, whereas the dimension-by-dimension approach requires tensor product grids. Numerical examples using both approaches demonstrate that the two-dimensional iterative adaptive radial basis function method yields accurate results.