A new grey forecasting model based on BP neural network and Markov chain

A new grey forecasting model based on BP neural network and Markov chain was proposed. In order to combine the grey forecasting model with neural network, an important theorem that the grey differential equation is equivalent to the time response model, was proved by analyzing the features of grey forecasting model（GM（1,1））. Based on this, the differential equation parameters were included in the network when the BP neural network was constructed, and the neural network was trained by extracting samples from grey system＇s known data. When BP network was converged, the whitened grey differential equation parameters were extracted and then the grey neural network forecasting model （GNNM（1,1）） was built. In order to reduce stochastic phenomenon in GNNM（1,1）, the state transition probability between two states was defined and the Markov transition matrix was established by building the residual sequences between grey forecasting and actual value. Thus, the new grey forecasting model（MNNGM（1,1）） was proposed by combining Markov chain with GNNM（1,1）. Based on the above discussion, three different approaches were put forward for forecasting China electricity demands. By comparing GM（1, 1） and GNNM（1,1） with the proposed model, the results indicate that the absolute mean error of MNNGM（1,1） is about 0.4 times of GNNM（1,1） and 0.2 times of GM（I, 1）, and the mean square error of MNNGM（1,1） is about 0.25 times of GNNM（1,1） and 0.1 times of GM（1,1）.

A Rub-Impact Recognition Method Based on Improved Convolutional Neural Network

Based on the theory of modal acoustic emission(AE),when the convolutional neural network(CNN)is used to identify rotor rub-impact faults,the training data has a small sample size,and the AE sound segment belongs to a single channel signal with less pixel-level information and strong local correlation.Due to the convolutional pooling operations of CNN,coarse-grained and edge information are lost,and the top-level information dimension in CNN network is low,which can easily lead to overfitting.To solve the above problems,we first propose the use of sound spectrograms and their differential features to construct multi-channel image input features suitable for CNN and fully exploit the intrinsic characteristics of the sound spectra.Then,the traditional CNN network structure is improved,and the outputs of all convolutional layers are connected as one layer constitutes a fused feature that contains information at each layer,and is input into the network’s fully connected layer for classification and identification.Experiments indicate that the improved CNN recognition algorithm has significantly improved recognition rate compared with CNN and dynamical neural network(DNN)algorithms.

Study of Synthesis Identification in Cutting Process with Fuzzy Neural Network

With the development of industrial production modernization, FMS and CIMS will become more and more popularized. For its control system is increasingly modeled, intellectualized and automatized, in order to raise the reliability and stability in the manufacturing process, the comprehensive monitoring and diagnosis aimed at cutting tool wear and chatter become more and more important and get rapid development. The paper tried to discuss of the intellectual status identification method based on acoustics-vibra characteristics of machining process, and propose that the working conditions may be taken as a core, complex fuzzy inference neural network model based on artificial neural network theory, and by using various kinds of modernized signal processing method to abstract enough characteristics parameters which will reflect overall processing status from machining acoustics-vibra signal as information source, to identify different working condition, and provide guarantee for automation and intelligence in machining process. The complex network is composed of NNw and NNs, Each of them is composed of BP model network, NNw is weight network at rule condition, NNs is decision-making network of each status. Y out is final inference result which is to take subordinate degree as weight from NNw, to weight reflecting result from NNs and obtain status inference of monitoring system. In the process of machining, the acoustics-vibor signal were gotten by the acoustimeter and the acceleration piezoelectricity detector, the date is analysed by the signal processing software in time and frequency domain, then form multi feature parameter vector of criterion pattern samples for the different stage of cutting chatter and acoustics-vibra multi feature parameter vector. The vector can give a accurate and comprehensive description for the cutting process, and have the characteristic which are speediness of time domain and veracity of frequency domain. The research works have been practically applied in identification of tool wear, cutting chatter, experiment results showed that it is practicable to identify the cutting chatter based on fuzzy neural network, and the new method based on fuzzy neural network can be applied to other state identification in machining process.

Completeness Problem of the Deep Neural Networks

Hornik, Stinchcombe & White have shown that the multilayer feed forward networks with enough hidden layers are universal approximators. Roux & Bengio have proved that adding hidden units yield a strictly improved modeling power, and Restricted Boltzmann Machines (RBM) are universal approximators of discrete distributions. In this paper, we provide yet another proof. The advantage of this new proof is that it will lead to several new learning algorithms. We prove that the Deep Neural Networks implement an expansion and the expansion is complete. First, we briefly review the basic Boltzmann Machine and that the invariant distributions of the Boltzmann Machine generate Markov chains. We then review the θ-transformation and its completeness, i.e. any function can be expanded by θ-transformation. We further review ABM (Attrasoft Boltzmann Machine). The invariant distribution of the ABM is a θ-transformation;therefore, an ABM can simulate any distribution. We discuss how to convert an ABM into a Deep Neural Network. Finally, by establishing the equivalence between an ABM and the Deep Neural Network, we prove that the Deep Neural Network is complete.

Applying acoustic emission and neural network to classify wheat seeds from weed seeds

In the present study,an expert weed seeds recognition system combining acoustic emissions analysis,Multilayer Feedforward Neural Network(MFNN)classifier was developed and tested for classifying wheat seeds.This experiment was performed for classifying two major important wheat varieties from five species of weed seeds.In order to produce sound signals,a 60o inclined glass plate was used.Fast Fourier Transform(FFT),Phase and Power Spectral Density(PSD)of impact signals were calculated.All features of sound signals are computed via a 1024-point FFT.After feature generation,60%of data sets were used for training,20%for validation,and remaining samples were selected for testing.The optimized MFNN model was found to have 500-12-2 and 500-10-2 architectures for“101”and“Shiroodi”wheat varieties,respectively.The selection of the optimal model was based on the evaluation of mean square error(MSE)and correct separation rate(CSR).The CSR percentages for two wheat varieties were 100%.Considering the overall aspects of the results,it can be stated that the developed system was successful enough to correlate the acoustic features with wheat seed type.

AN ACCELERATION STRATEGY FOR RANDOMIZE-THEN-OPTIMIZE SAMPLING VIA DEEP NEURAL NETWORKS

Randomize-then-optimize (RTO) is widely used for sampling from posterior distributions in Bayesian inverse problems. However, RTO can be computationally intensive forcomplexity problems due to repetitive evaluations of the expensive forward model and itsgradient. In this work, we present a novel goal-oriented deep neural networks (DNN) surrogate approach to substantially reduce the computation burden of RTO. In particular,we propose to drawn the training points for the DNN-surrogate from a local approximatedposterior distribution – yielding a flexible and efficient sampling algorithm that convergesto the direct RTO approach. We present a Bayesian inverse problem governed by ellipticPDEs to demonstrate the computational accuracy and efficiency of our DNN-RTO approach, which shows that DNN-RTO can significantly outperform the traditional RTO.

Separable Shadow Hamiltonian Hybrid Monte Carlo for Bayesian Neural Network Inference in wind speed forecasting

Accurate wind speed and consequently wind power forecasts form a critical enabling tool for large scale wind energy adoption.Probabilistic machine learning models such as Bayesian Neural Network(BNN)models are often preferred in the forecasting task as they facilitate estimates of predictive uncertainty and automatic relevance determination(ARD).Hybrid Monte Carlo(HMC)is widely used to perform asymptotically exact inference of the network parameters.A significant limitation to the increased adoption of HMC in inference for large scale machine learning systems is the exponential degradation of the acceptance rates and the corresponding effective sample sizes with increasing model dimensionality due to numerical integration errors.This paper presents a solution to this problem by sampling from a modified or shadow Hamiltonian that is conserved to a higher-order by the leapfrog integrator.BNNs trained using Separable Shadow Hamiltonian Hybrid Monte Carlo(S2HMC)are used to forecast one hour ahead wind speeds on the Wind Atlas for South Africa(WASA)datasets.Experimental results find that S2HMC yields higher effective sample sizes than the competing HMC.The predictive performance of S2HMC and HMC based BNNs is found to be similar.We further perform hierarchical inference for BNN parameters by combining the S2HMC sampler with Gibbs sampling of hyperparameters for relevance determination.A generalisable ARD committee framework is introduced to synthesise the various sampler ARD outputs into robust feature selections.Experimental results show that this ARD committee approach selects features of high predictive information value.Further,the results show that dimensionality reduction performed through this approach improves the sampling performance of samplers that suffer from random walk behaviour such as Metropolis–Hastings(MH).

Machine auscultation: enabling machine diagnostics using convolutional neural networks and large-scale machine audio data

Acoustic signals play an essential role in machine state monitoring. Efficient processing of real-time machine acoustic signals improves production quality. However, generating semantically useful information from sound signals is an ill-defined problem that exhibits a highly non-linear relationship between sound and subjective perceptions. This paper outlines two neural network models to analyze and classify acoustic signals emanating from machines:(i) a backpropagation neural network (BPNN);and (ii) a convolutional neural network (CNN). Microphones are used to collect acoustic data for training models from a computer numeric control (CNC) lathe. Numerical experiments demonstrate that CNN performs better than the BP-NN.

基于神经网络和Markov链的交通流实时滚动预测

将神经网络与Markov链理论应用于随机波动的交通流预测,提出一种交通流实时滚动预测方法TDFNM。该方法采用BP网络构建交通流基准预测曲线,使用SOM网络划分残差的Markov链状态,计算各状态加权中心及状态转移概率矩阵,以此预测未来状态,并以加权中点修正计算得到精度较高的预测值,同时实现实时滚动预测。采用方法TDFNM对实测交通流量进行仿真实验,结果表明,该方法比常规BP网络具有更高的准确性,而且具有较强的适应性。

基于神经网络—Markov状态模型的电力电量预测

提出了神经网络-Markov状态预测模型,采用误差修正的方法,对电力电量进行预测。我们采用1998年1月-2009年7月的数据对全国发电总量进行预测。结果表明:神经网络-Markov状态预测模型明显提高了预测精度,说明该模型对电力电量的预测更为有效。

基于Markov-BP神经网络的武汉市物流需求预测

物流需求预测是城市发展规划中的重要组成部分,为了能够科学地预测出武汉市的物流需求,选择武汉市地区生产总值、社会商品零售总值及货物进出口作为输入指标,将货物运输量作为输出指标,利用BP神经网络模型进行预测。在此基础上,借助马尔可夫链(Markov)对误差值进行修正,使平均相对误差从7.3%下降至1.9%。结果表明,与单一的BP神经网络模型以及其他神经网络组合方法相比,Markov-BP神经网络模型的预测精度更高,使用Markov-BP神经网络模型,对武汉市未来物流需求预测具有一定的参考价值。

基于BP-Markov模型的技术创新融资风险耦合分析

为进一步区分不同技术创新措施的风险形成机制,文章利用Markov链式遗传优化方法,构建了一个基于专家意见评估的紧逼型耦合BP精读训练模式,以此进行面向技术创新融资风险的耦合BP-Markov改进检验。结果显示,经风险规避行为叠合剔除后验证的指标在技术创新能力、政府支持方面表现出对融资的阻滞作用,而技术创新产权所表征的知识产权纠纷成为影响创新主体顺利获得外部融资的关键因素。

融合Markov与BP神经网络的纯电动汽车销售量预测研究

随着纯电动汽车市场的不断扩大,对销售量的精确预测成为人们当前关注的热点。提取影响销售量的7个关键因素以及2017~2019年9月的纯电动汽车销售量,首先利用BP神经网络模型对33个月的数据进行测试,并用训练好的模型预测2019年1~9月销售量,再利用马尔科夫(Markov)模型将BP神经网络模型预测的相对误差划分为6种状态,对预测结果进行修正。通过对BP神经网络模型与Markov-BP神经网络模型预测结果进行对比检验,发现Markov-BP神经网络的预测准确度更高,表明采用Markov-BP神经网络模型对纯电动汽车月度销售量进行预测具有一定现实意义。

Training Kohonen Networks by Using an Improved Genetic Algorithm

ＴｒａｉｎｉｎｇＫｏｈｏｎｅｎＮｅｔｗｏｒｋｓｂｙＵｓｉｎｇａｎＩｍｐｒｏｖｅｄＧｅｎｅｔｉｃＡｌｇｏｒｉｔｈｍＳｏｎｇＡｉｇｕｏ（宋爱国）ＬｕＪｉｒｅｎ（陆佶人）（ＤｅｐａｒｔｍｅｎｔｏｆＲａｄｉｏＥｎｇｉｎｅｅｒｉｎｇ，ＳｏｕｔｈｅａｓｔＵｎｉ...

基于Markov-Elman神经网络的消费者信心指数预测模型构建

以2007年1月至2018年9月我国消费者信心指数的月度数据为样本,分别构建马尔科夫链和elman神经网络,拟合并预测消费者信心指数的未来走势。结果表明:马尔科夫链和elman神经网络的预测效果良好、模型预测精度高、误差小,且两种方法的预测结果一致,为消费者信心指数及同类指数的预测提供一种新的思路和方法。

基于迁移学习和CNN-LSTM的水轮机空化状态识别方法

针对水轮机空化声发射信号中包含较多噪声、依赖人工降噪与特征提取以及深度学习模型准确率极度依赖海量训练数据的问题,提出一种基于迁移学习和卷积神经网络-长短时记忆网络(CNN-LSTM)的水轮机空化状态识别方法。首先,将数据输入CNN中提取隐含特征;然后,在LSTM中提取特征隐含的时序信息并输出空化类型,通过训练网络参数建立基于CNN-LSTM的空化状态识别模型;最后,引入迁移学习对类似工况进行空化状态识别。结果表明:该模型能准确识别出3种不同的水轮机空化类型,其平均识别准确率达到较高水平;与传统深度学习模型相比,该模型在极少样本学习任务中的识别准确率具有明显优势。

试验环境水下声信号的特征提取方法

水下试验环境参数的反演是水声学研究领域的重要内容。而当前研究的关键是通过对水下声信号做特征提取从而获取参数信息。针对特征提取较难、模型很难拟合等问题。本文提出了一种试验环境水下声信号的特征提取方法。将水下声信号同时用梅尔频谱倒谱系数及线性预测系数处理,两者运用特征加权组合方法得到新的特征矩阵;再应用映射插值算法对特征矩阵进行处理,获得适应神经网络输入的三通道矩阵。本文选取的网络模型为残差神经网络。利用实验室所录制的对河口水库数据集测试表明,本文提出的特征提取方法普遍优于仅利用梅尔频谱倒谱系数或线性预测系数的特征处理方法。利用单频矩形脉冲信号对环境进行深度5分类,准确率平均提升2%。利用线性调频信号对环境进行深度5分类,准确率平均提升2.03%。本文提出的特征提取方法对线性调频信号在深度分类任务下处理的结果要优于单频矩形脉冲信号处理的结果。

基于神经网络与改进马尔可夫链中压背景噪声建模研究

在中压电力线通信中,信道噪声构成复杂,需根据不同类型噪声单独分析建模。针对一段特定中压线路的背景噪声,提出了一种基于小波包变换的噪声模型,将得到的小波包系数分别进行神经网络训练和改进马尔可夫链计算转移概率矩阵,得到新的小波包系数重构噪声信号,并进行仿真验证及去噪,同时将2种方法与传统直接神经网络训练比较分析。结果表明,基于改进马尔可夫链方法所建噪声比传统马尔可夫链方法更加准确,基于小波包变换的神经网络方法所建噪声与原噪声相似度更高,去噪效果更好,且优于传统神经网络训练方法,为进一步研究中压电力线通信提供了可行性方案。

基于GRNN-MC的变压器振动信号预测

变压器振动信号是评估其工作状态的重要参数之一,与绕组松动或变形等隐患密切相关,为揭示变压器振动信号的变化趋势,本文提出了一种基于广义回归神经网络和马尔科夫链修正的变压器振动信号预测方法。即分别以变压器运行电压、负载电流和振动信号归一化特征频率为输入和输出建立变压器振动信号广义回归神经网络预测模型,然后引入马尔科夫链并结合负载电流的变化对振动信号计算结果进行修正以获得最终的预测结果。对某500 kV变压器振动在线监测信号的分析结果表明:经马尔科夫链修正后的变压器广义回归神经网络振动信号预测模型预测精度高,可为变压器绕组运行状态的振动监测技术提供重要参考。

基于CNN和MFCC的供水管网漏损声信号识别方法

针对供水管网漏损识别效率低和对人工经验依赖性强等问题,基于卷积神经网络(CNN)和梅尔频率倒谱系数(MFCC)提出了一种供水管网漏损声信号识别方法。对噪声记录仪和水音传感器采集的漏损声信号提取MFCC及其一、二阶差分作为漏损声信号特征,得到了包含漏损特征的特征图像,将其输入到CNN模型,通过超参数优化后最终得到了漏损识别模型。结果表明,使用MFCC与MFCC的一阶差分特征参数组合作为输入特征训练模型时的识别效果最好,其测试集准确率达到95.26%,F1分数达到89.22%,具备优良的漏损识别能力。