An electric energy consumer characterization framework based on data mining techniques

This paper presents an electricity consumer characterization framework based on a knowledge discovery in databases (KDD) procedure, supported by data mining (DM) techniques, applied on the different stages of the process. The core of this framework is a data mining model based on a combination of unsupervised and supervised learning techniques. Two main modules compose this framework: the load profiling module and the classification module. The load profiling module creates a set of consumer classes using a clustering operation and the representative load profiles for each class. The classification module uses this knowledge to build a classification model able to assign different consumers to the existing classes. The quality of this framework is illustrated with a case study concerning a real database of LV consumers from the Portuguese distribution company.     

Biosignal pattern recognition and interpretation systems

A general framework is given to describe pattern recognition and interpretation. Pattern analysis stages are described, with consideration of difficulties in implementation and uncertainties present at each level. The main forms of pattern analysis-statistical, syntactic, and artificial intelligence (connectionist and symbolic) methods-have different strengths and weaknesses, depending on the stage of pattern analysis at which they are used. In general, statistical, syntactic, and connectionist techniques are used for pattern recognition, and statistical and symbolic techniques are used for pattern interpretation. Largely, pattern interpretation involves reasoning with uncertainty. Multichannel recordings increase the information available about specific physiologic events, at the expense of processing complexity     

基于多层极限学习机的电能质量扰动多标签分类算法

电力系统中电能质量扰动分类特征选择标准不统一、泛化能力差、分类效果与分类效率有待提高。为了解决这些问题,一方面,引入多层极限学习机自编码器,优化输入权重,完成电能质量扰动信号的特征提取。另一方面,引入多标签排位分类算法,充分考虑各标签之间的相关性,完成电能质量扰动的分类。基于两种算法,设计出基于多层极限学习机的多标签分类模型,并得到多层极限学习机的最优网络结构和多标签分类的最佳分类阈值。实验结果表明,所提方法适用于电能质量单一扰动和复合扰动的分类,改善了分类效果和分类效率,具有较高的分类精度、良好的抗噪能力和泛化能力。     

深度学习在智能电网中的应用现状分析与展望

深度学习是机器学习研究中的一个新领域,其强大的数据分析、预测、分类能力契合智能电网中大数据应用的需求。文中首先总结了深度学习基本思想,介绍深度学习的5种模型(生成式对抗网络、递归神经网络、卷积神经网络、堆叠自动编码器和深度信念网络)的结构、基本原理、训练方法,概括其应用特征。综述了电力系统中的故障诊断、暂态稳定性分析、负荷及新能源功率预测、运行调控等应用深度学习技术的研究现状。针对深度学习的技术特点,结合电力系统各生产环节,构建深度学习技术在电力系统中的应用框架。最后,从多能源系统运行调控、电力电子化系统安全分析、柔性设备故障诊断、电力信息物理系统的安全防护等方面对深度学习应用进行展望。     

Pattern classification of time‐series EMG signals using neural networks

This paper proposes a pattern classification method of time‐series EMG signals for prosthetic control. To achieve successful classification for non‐stationary EMG signals, a new neural network structure that combines a common back‐propagation neural network with recurrent neural filters is used. A convergence time of the network learning can be regulated by a new learning method based on dynamics of a terminal attractor. The experiments of pattern classification and prosthetic control are carried out for several subjects including an amputee. It is shown from the results that the proposed method improves learning/classification ability for stationary and non‐stationary EMG signals during a series of continuous motions. Copyright © 2000 John Wiley & Sons, Ltd.     

Fruit juice–alcohol mixture analysis using machine learning and electronic nose

The aim of this study is to analyze the raw data collected from a fruit juice–alcohol mixture (a fruit juice–alcohol mixture and a fruit juice–multiple alcohol mixture) and the Halal authentication of a fruit juice–alcohol mixture with electronic nose. Machine learning techniques such as naïve Bayesian classifier, K‐nearest neighbors (K‐NN), linear discriminant analysis (LDA), decision tree, artificial neural network (ANN), and support vector machine (SVM) were used to classify the feature of these collected raw data. There are three types of classification: the first one is a fruit juice and an alcohol mixture type; the second is a fruit juice and multiple alcohol mixture types, and the third is a Halal authentication of a fruit juice and alcohol mixture. We aimed at making cocktails with more successful results on the first two types of classification in the work. Also, we focused on Halal authentication of fruit juice–alcohol mixture in the third classification. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

CNN 集成机器学习的金属缺陷少样本分类方法

针对金属缺陷分类,以深度学习为代表的分类方法主要是基于大规模数据的统计学习方法,一方面需要大量优质的标注样本,另一方面对数据中未能涵盖的样本泛化性能差。提出了一种利用集成学习思想,将人类分类知识嵌入到深度学习的少样本分类方法。首先搭建了一个卷积神经网络作为分类模型的骨干网络,并设计了一个利用机器学习的类人学习模块,利用人类分类所用特征进行分类。此外,为了提高模型的泛化性、鲁棒性和更好的融合效果,设计了一种以对数函数为核心的数学集成模型,模块中的数学集成模型利用集成学习思想将骨干网络和类人学习模块的输出进行耦合。实验结果表明,对于小训练集大测试集的金属缺陷数据在分类性能和训练参数量方面优于深度学习方法。此外,类人学习模块和数学集成模型嵌入到不同的骨干网络上均取得了很好的性能,表明所提出的方法适用于多种深度卷积神经网络。     

基于阶段聚焦损失和并行增广策略的 遥感图像场景分类

随着深度学习的不断普及,卷积神经网络已成为遥感图像场景分类的主要手段,然而当前的研究主要集中于多网络主干的信息融合以及注意力机制等领域,在提高分类精度的同时也带来更高的计算复杂度。针对上述问题,分别从改进卷积神经网络损失函数和设计新的样本训练策略两个角度出发,在不增加计算复杂度的前提下,提升卷积神经网络的分类性能。首先,在对传统交叉熵和Focal loss损失函数进行分析的基础上,提出一种阶段聚焦损失函数,该损失函数可以在训练阶段对卷积网络进行有侧重的性能挖掘。其次,设计了一种并行样本训练策略,将采用Gridmask算法增广后的样本图像和原始样本图像,分为两路输入卷积网络进行并行训练,进一步提升卷积网络的分类性能。实验结果表明,所提出的算法分别在AID和NWPU-RESISC45两个大规模数据库上取得了96.72%和93.95%的检测精度,可以显著提升遥感图像场景分类的性能。     

Pattern analysis and classification for security evaluation in power networks

Security evaluation is a major concern in real time operation of electric power networks, exhibiting behavioral patterns under abnormal conditions. Security assessment and evaluation can be viewed as a pattern analysis task identifying abnormal patterns of the power system behavior under highly loaded conditions. Traditional method of security evaluation are highly time consuming and infeasible for direct on-line implementation. This paper presents application of pattern directed inference system for static and transient security evaluation and classification. A straightforward and quick procedure called Sequential Forward Selection method is used for feature selection process. The classifier model in the pattern directed inference system is designed using different pattern classifier algorithms, viz., conventional, neural network and machine learning classifiers. Support Vector Machine (SVM), one of the popular machine learning classifier, is recognized as a suitable pattern classifier for security evaluation problem. The generalization performance of SVM classifier is greatly influenced by the proper setting of its parameters. This paper also addresses different heuristic optimization techniques used in the selection of SVM parameters. The design, development and performance of different classifiers for power system security classification are presented in detail. Simulation work is performed on standard New England 39-bus benchmark system and the feasibility of implementation of the proposed SVM based classifier system for on-line security evaluation is also discussed.     

Determining hearing threshold from brain stem evoked potentials.Optimizing a neural network to improve classification performance

Feed-forward neural networks in conjunction with back-propagation are an effective tool to automate the classification of biomedical signals. Most of the neural network research to date has been done with a view to accelerate learning speed. In the medical context, however, generalisation may be more important than learning speed. With the brain stem auditory evoked potential classification task described in this study, the authors found that parameter values that gave fastest learning could result in poor generalisation. In order to achieve maximum generalisation, it was necessary to fine tune the neural net for gain, momentum, batch size, and hidden layer size. Although this maximization could be time consuming, especially with larger training sets, the authors' results suggest that fine tuning parameters can have important clinical consequences, which justifies the time involved. In the authors' case, fine tuning parameters for high generalisation had the additional effect of reducing false negative classifications, with only a small sacrifice in learning speed     

适用于海量负荷数据分类的高性能反向传播神经网络算法

负荷分类对于指导电网发用电规划与保证电网可靠运行具有重要意义。面向负荷数据海量化与复杂化趋势,传统负荷分类方法已无法满足用电大数据分析要求。首先,针对用户侧数据体量大、类型多、速度快等特点,在Spark平台上将反向传播神经网络(BPNN)算法并行化,实现对海量负荷数据的高效分类。然后,通过对训练样本抽样分块以降低各网络学习时间,针对分布式后BPNN基分类器由于学习样本缺失潜在的准确度下降问题,采用集成学习予以改善。并通过BPNN学习不同训练样本块构建差异化基分类器,对基分类结果多数投票得到最终分类结果。另外,提供了一种基于K-means和K-medoids聚类的负荷数据训练样本选取方法。算例表明所提方法既能对负荷曲线有效分类,又能大幅提高海量数据的处理效率。     

一种基于DBN的高光谱遥感图像分类方法

高光谱遥感数据越来越普及并为人们广泛使用,基于高光谱数据的地面物体精确分类是高光谱遥感技术的核心应用之一。对高光谱数据进行提特征提取是进行地物分类的有效方法。深度学习是机器学习研究中的新领域,它多隐层的多层感知器结构使其能够学习到对数据有更本质的刻画的特征,在图像分类和可视化领域取得了更好的成绩。深度置信网（deep belief network ,DBN）是深度学习网络中常见的模型。利用高光谱数据的高维特性,搭建基于DBN的高光谱图像分类模型,结合高光谱数据的空间结构对地物进行分类。实验表明,基于DBN的高光谱图像分类方法可以得到更好的分类效果。     

Power Utility Nontechnical Loss Analysis With Extreme Learning Machine Method

This paper presents a new approach to nontechnical loss (NTL) analysis for utilities using the modern computational technique extreme learning machine (ELM). Nontechnical losses represent a significant proportion of electricity losses in both developing and developed countries. The ELM-based approach presented here uses customer load-profile information to expose abnormal behavior that is known to be highly correlated with NTL activities. This approach provides a method of data mining for this purpose, and it involves extracting patterns of customer behavior from historical kWh consumption data. The results yield classification classes that are used to reveal whether any significant behavior that emerges is due to irregularities in consumption. In this paper, ELM and online sequential-ELM (OS-ELM) algorithms are both used to achieve an improved classification performance and to increase accuracy of results. A comparison of this approach with other classification techniques, such as the support vector machine (SVM) algorithm, is also undertaken and the ELM performance and accuracy in NTL analysis is shown to be superior.     

PCA和Elman网络在移动学习策略分类中的应用

针对传统的大学生英语移动学习策略分类方法准确率较低的情况,提出了一种主成分分析（PCA）和Elman神经网络相结合的分类模型。首先,用PCA对所获得的移动学习策略原始数据作数据降维处理,提取前5个主成分,建立新的特征样本矩阵,再对Elman神经网络进行训练和泛化能力测试。仿真结果表明：单一的BPNN分类准确率为70.0%,单一的Elman网络分类准确率为80.0%,PCA Elman网络分类准确率为100.0%,PCA Elman网络模型简化了单一Elman网络的结构,提高了网络的训练速率、分类准确率和泛化能力,验证了所提出的模型的有效性。     

基于LVQ神经网络的变压器故障诊断方法

结合模糊理论,提出一种基于学习向量量化器(LVQ)的变压器故障诊断方法。它首先在无监督学习模式下,采用数据压缩技术,完成输入空间上的向量重构。接着结合监督学习机制,从输入数据选择特征赋予每个类。该方法是一种将自组织映射(SOM)和监督学习模式结合起来的自适应模式分类技术,具有结构简单,适应性强和分类精度高的特点。变压器故障诊断实例显示了该方法的有效性。     

The BCI competition. III: Validating alternative approaches to actual BCI problems.

A brain-computer interface (BCI) is a system that allows its users to control external devices with brain activity. Although the proof-of-concept was given decades ago, the reliable translation of user intent into device control commands is still a major challenge. Success requires the effective interaction of two adaptive controllers: the user's brain, which produces brain activity that encodes intent, and the BCI system, which translates that activity into device control commands. In order to facilitate this interaction, many laboratories are exploring a variety of signal analysis techniques to improve the adaptation of the BCI system to the user. In the literature, many machine learning and pattern classification algorithms have been reported to give impressive results when applied to BCI data in offline analyses. However, it is more difficult to evaluate their relative value for actual online use. BCI data competitions have been organized to provide objective formal evaluations of alternative methods. Prompted by the great interest in the first two BCI Competitions, we organized the third BCI Competition to address several of the most difficult and important analysis problems in BCI research. The paper describes the data sets that were provided to the competitors and gives an overview of the results.     

基于预训练BERT模型的客服工单自动分类研究

本文介绍了工单文本分类的理论和应用,并对文本分析的分词、机器学习、深度学习等技术方法进行了描述。基于预训练BERT模型提出了95598客服工单自动分类的方法,设计了电力客服工单自动分类的流程,最后通过一个实际的案例对算法模型进行校验,并与传统的文本挖掘方法进行了对比。算例的结果表明,所使用的工单分类算法能显著提高分类的准确性,在分类效率上也较高。     

Fuzzy-neuro approach to fault classification for transmission lineprotection

This paper presents a new approach to real-time fault detection and classification in power transmission systems by using fuzzy-neuro techniques. The integration with neural network technology enhances fuzzy logic systems on learning capabilities. The symmetrical components in combination with three line currents are utilized to detect fault types such as single line-to ground, line-to-line, double line-to-ground and three line-to-ground, and then to define the faulty line. Computer simulation results are shown in this paper and they indicate this approach can be used as an effective tool for high speed digital relaying, as the correct detection is achieved in less than 10 ms     

面向图像语义分类的视觉单词集成学习方法

针对图像的特性,提出了1种视觉单词集成学习方法.该方法建立在3种初始映射方法的基础上,并充分利用图像的矩、纹理直方图、图像傅里叶描述子等图像视觉特征来分类图像语义.相对于3种初始映射方法,采用Boosting集成学习方法生成的视觉单词集合在图像语义分类上比单独使用任意1种映射方法生成的视觉单词集合有明显的提高.实验结果表明,本文所提出的视觉单词Boosting集成学习方法在图像语义区分性和描述能力方面是有效的,能充分反映人对图像内容的理解,具有很好的应用价值.