基于神经网络的齿轮箱故障诊断专家系统及应用

在机械故障诊断中,故障与征兆间并非都是一一对应或线性关系。机械设备的故障诊断实质上是一种模式识别,诊断结果的准确度往往取决于模式识别的精度。传统的故障诊断方法基于统计模式识别,在识别速度、可靠性等方面有一定的局限性。目前,基于智能模式识别的故障诊断技术(如诊断专家系统)有了很大的发展,但在知识获取、自学习等方面有待继续发     

方向基函数神经网络在说话人识别中的应用

首先对方向基函数神经网络模型(Direction basisFunctionNeuralNetworks)进行了描述。这种模型可以通过计算矢量间的夹角来对矢量进行相似性判别。与径向基函数神经网络模型(Radial basisFunctionNeuralNetworks)不同的是,它特别适合于具有方向不变性的模式识别领域,如语音识别。重点探讨了方向基函数神经网络在说话人识别中的应用,在此基础上研制成功了1个说话人确认系统并用硬件进行了实现。实验数据表明这个系统具有很好的性能。     

基于深度学习理论的风力机结冰预警模型

基于深度学习理论,利用风电机组运行数据训练深度神经网络模型,用以表征信号与结冰状态之间复杂的映射关系。试验结果表明该模型实现了多工况、大样本下的结冰特征量提取与结冰状态的识别,具有较高的精度,为风电机组结冰状态的预测提供了新的思路。     

遗传算法优化的小波神经网络在短期电力负荷预测中的应用

本文利用遗产算法的全局优化搜索能力来优化了小波神经网络,建立了基于遗产算法的小波神经网络短期电力负荷预测模型,克服了BP神经网络自身算法的缺陷,得到了更高的学习精度和更快的收敛速度,经实例也验证了该模型能有效地提高预测精度,减小负荷预测误差,避免了BP神经网络的固有缺陷.     

一种基于感受野增强的人脸检测方法

为了解决多任务级联卷积神经网络（MTCNN）算法网络模型在小人脸检测方面鲁棒性较低的问题,提出了一种基于感受野增强的网络模型。首先,为MTCNN算法模型中的R-Net网络和O-Net网络添加感受野模块（receptive field blocks,RFB-S）。其次,通过添加批量标准化和全局平均池化,加速网络模型的收敛,防止模型过拟合。最后,调整网络任务的权重,P-Net和R-Net网络用于人脸区域粗筛选,O-Net网络用于人脸区域精筛选以及人脸关键点回归。实验结果表明,与MTCNN算法网络模型相比,所提模型缩小了16%,但检测速度提升了9%,在FDDB数据集上的检测精度提高了2.3%。因此,基于感受野增强的网络模型能有效完成人脸的检测任务,增强对小人脸检测的鲁棒性,可为人脸识别、表情识别等提供技术支持。     

基于进化神经网络重型汽车ABS故障诊断研究

通过分析重型汽车的ABS制动系统,研究其系统结构。对进化神经网络算法进行分析,研究了算法实现的基本流程和方法。通过进行学习样本数据进行了收集,得到50组样本数据。运用Matlap软件对50组样本数据中的45组样本数据进行了学习训练,得到了进化神经网络重型汽车故障诊断模型。通过检测样本对故障诊断模型进行了检验,证明了该模型有较高精度的输出值。     

神经网络法识别抽油机井示功图的研究及应用

示功图是判断油井生产状况的重要依据。神经网络能够反映任意非线性的映射关系,从而可以应用于图形识别。本文主要依据BP神经网络判定示功图类型的实现过程,阐述了BP神经网络的基本原理,建立了模式识别系统,并给出了部分应用实例。     

动态神经网络方法在车内噪声参考信号拾取上的应用

推导了用于实时控制的动态神经网络数值方法,并应用该方法通过测量关键点的振动加速度信号来识别车内噪声信号,该法既避免了传统自适应有源消声中的声反馈问题,又具有实时性。实验研究表明,由于动态网络具有良好的实时跟踪特性,即使在工况发生改变时,网络也可在短时间内迅速完成跟踪。     

基于改进小波神经网络模型的汤河含沙量预测研究

本文引入变量权重系数对小波神经网络进行改进,改善传统小波神经网络模型易出现局部收敛的缺陷,并将改进小波神经网络模型对汤河含沙量进行预测。研究结果表明:改进的小波神经网络模型改进了传统BP神经网络模型存在局部收敛的缺陷,在河流含沙量预测中,模拟的含沙量相对误差符合含沙量预测规范精度,可用于河流含沙量预测。研究成果对于河流含沙量预测提供参考价值。     

故障诊断神经网络的发展与研究

人工神经网络或神经网络又可叫连接网络模型,并行分布处理模型及神经组织系统。它是依据现有对人脑思维过程研究和掌握的基础,用数学办法对其进行简化、抽象和模拟,并反映人脑功能的基本特性。 近年来,由于计算机、现代测置技术和信号处理技术的迅速发展,机械故障诊断技术取得了很大的进展。人们已开发和研究了一些较成熟和完善的诊断技术和理论方法,如铁谱分析、声发射、红外测温、油液分析和各种无损监测等技术及信号处理、模式识别、专家系统和模糊识别等理论方法。应用这些技术手段和理论方法,人们可对在多种工作环境条件及运行状态下的机器式工程系统的许多故障模式进行监测、识别、诊断和排除。     

基于轻量级卷积神经网络的苹果表面缺陷检测方法

为了解决目前农业信息领域对苹果表面缺陷检测准确率低的问题,提出一种基于轻量级卷积神经网络的苹果表面缺陷检测方法.首先采集苹果缺陷样本图片制作实验数据集用于模型训练和测试;其次在AlexNet网络结构的基础上,引入深度可分离卷积代替原有网络中的标准卷积运算来进行图像特征的提取;最后利用全局平均池化方法代替原有网络中的全连...     

模式识别中的光字符识别技术及应用综述

介绍了模式识别的概念,阐述了模式识别中的统计决策法、结构模式识别、模糊模式识别与人工神经网络识别,重点介绍了字符识别中的光字符识别技术。OCR技术综合了数字图像处理、计算机图形学和人工智能等多方面的知识,成为当今模式识别领域最活跃的研究内容之一。OCR技术在文字识别中有着广泛应用,文中对OCR技术在我国汉字印刷体和手写体识别中的应用作了重点介绍,最后扼要介绍了OCR技术的研究方向。     

Flagging cartel participants with deep learning based on convolutional neural networks

Adding to the literature on the data-driven detection of bid-rigging cartels, we propose a novel approach based on deep learning (a subfield of artificial intelligence) that flags cartel participants based on their pairwise bidding interactions with other firms. More concisely, we combine a so-called convolutional neural network for image recognition with graphs that in a pairwise manner plot the normalized bids of some reference firm against the normalized bids of any other firms participating in the same tenders as the reference firm. Based on Japanese and Swiss procurement data, we construct such graphs for both collusive and competitive episodes (i.e when a bid-rigging cartel is or is not active) and we use a subset of graphs to train the neural network such that it learns distinguishing collusive from competitive bidding patterns. With the remaining graphs, we test the neural network’s out-of-sample performance in correctly classifying collusive and competitive bidding interactions. We obtain a very decent average accuracy of around 95% or slightly higher when either applying the method within Japanese, Swiss, or mixed data (in which Swiss and Japanese graphs are pooled). When using data from one country for training to test the trained model’s performance in the other country (i.e. transnationally), predictive performance decreases (likely due to institutional differences in procurement procedures across countries), but often remains satisfactorily high. All in all, the generally quite high accuracy of the convolutional neural network despite being trained in a rather small sample of a few 100 graphs points to a large potential of deep learning approaches for flagging and fighting bid-rigging cartels.     

Quantized Hopfield networks and tabu search for manufacturing cell formation problems

The use of neural networks in the design of cellular manufacturing system is not new. This paper presents an application of modified Hopfield neural networks in order to solve cell formation problems: the quantized and fluctuated Hopfield neural networks (QFHN). This kind of Hopfield network combined with the “tabu search” approach were primarily used in a hybrid procedure in order to solve the cell formation for big sizes industrial data set. The problem is formulated as a 0/1 linear and integer programming model in order to minimize the dissimilarities between machines and/or parts. Our hybrid approach allows us to obtain optimal or nearly optimal solutions very frequently and much more quickly than traditional Hopfield networks. It is also illustrated that the fluctuation associated with this quantization may enable the network to escape from local minima, to converge to global minima, and consequently to obtain optimal solutions very frequently and much more quickly than pure quantized Hopfield networks (QHN). The effectiveness of the proposed approach is flexibility it gives us, for example, in time problem-solving for large-scale and speed of execution when we apply it.     

Fashion retail forecasting by evolutionary neural networks

Recent literature on nonlinear models has shown that neural networks are versatile tools for forecasting. However, the search for an ideal network structure is a complex task. Evolutionary computation is a promising global search approach for feature and model selection. In this paper, an evolutionary computation approach is proposed in searching for the ideal network structure for a forecasting system. Two years’ apparel sales data are used in the analysis. The optimized neural networks structure for the forecasting of apparel sales is developed. The performances of the models are compared with the basic fully connected neural networks and the traditional forecasting models. We find that the proposed algorithms are useful for fashion retail forecasting, and the performance of it is better than the traditional SARIMA model for products with features of low demand uncertainty and weak seasonal trends. It is applicable for fashion retailers to produce short-term retail forecasting for apparels, which share these features.     

基于MBP算法和深度学习的人脸识别

为了解决在深度学习提取人脸图像特征时,易忽略其局部结构特征和缺乏对其旋转不变性学习的问题,提出了一种基于单演局部二值模式(MBP)与深度学习相结合的高效率人脸识别方法。首先,用多尺度单演滤波器对图像进行滤波,得到幅值和方向信息;其次,用LBP算法和象限比特的方法进行编码,分块计算组合其直方图特征;然后,将提取的单演特征作为深度信念网络(DBN)的输入,逐层训练优化网络参数,得到优异的网络模型;最后,将训练好的DBN网络在ORL人脸数据库上进行人脸识别实验,进行识别率计算,其识别率为98.75%。所提出的方法使用无监督的贪婪算法,隐藏层设定为2层,使用反向传播算法优化网络。相较于已知的人脸识别方法,MBP+DBN算法对光照、表情和部分遮挡变化具有较好的鲁棒性,在人脸识别中识别率较高,具有一定的优势,为图像特征提供了一种新的识别方法。     

The use of ICTs and income distribution in Brazil: A machine learning explanation using SHAP values

This study explores the complex relationship between information and communication technologies (ICTs) and socioeconomic characteristics. We employ a cutting-edge explainable machine learning approach, known as SHAP values, to interpret an XGBoost and neural network model, as well as benchmark traditional econometric methods. The application of machine learning algorithms combined with the SHAP methodology reveals complex nonlinear relationships in the data and important insights to guide tailored policy-making. Our results suggest that there is an interaction between education and ICTs that contributes to income prediction. Furthermore, level of education and age are found to be positively associated with income, while gender presents a negative relationship; that is, women earn less than men on average. This study highlights the need for more efficient public policies to fight gender inequality in Brazil. It is also important to introduce policies that promote quality education and the teaching of skills related to technology and digitalization to prepare individuals for changes in the job market and avoid the digital divide and increasing social inequality.     

基于图像的地铁车站施工人员不安全行为识别研究

为了解决人工监测不能实时看护和及时管理地铁车站施工人员的不安全行为等问题,构建了基于KNN,MLP和LSTM模型的不安全行为识别神经网络模型。首先,通过行为理论研究和现场调查分析,对地铁车站施工不安全行为进行了分类;其次,通过实验构建人体数据集,基于人体骨骼关节点提取不安全行为特征,并进行模型训练;最后,基于MobileNet V1的SSD目标检测算法对施工人员进行定位和追踪,结合射线法判断目标是否跨越不安全区域并发出警报,搭建神经网络模型对施工人员的不安全行为进行识别,并获得计算识别率。结果表明：传统机器学习算法KNN总体准确率为93.45%,优化后的MLP和LSTM两种神经网络模型总体准确率分别达到93.94%和93.68%,相对KNN算法分别提高了0.49%和0.23%。因此所提模型能有效识别施工人员不安全行为,可为地铁施工安全智能识别技术应用提供参考。