Pseudo-label based semi-supervised learning in the distributed machine learning framework

With the emergence of various intelligent applications,machine learning technologies face lots of challenges including large-scale models,application oriented real-time dataset and limited capabilities of nodes in practice.Therefore,distributed machine learning(DML) and semi-supervised learning methods which help solve these problems have been addressed in both academia and industry.In this paper,the semi-supervised learning method and the data parallelism DML framework are combined.The pseudo-label based local loss function for each distributed node is studied,and the stochastic gradient descent(SGD) based distributed parameter update principle is derived.A demo that implements the pseudo-label based semi-supervised learning in the DML framework is conducted,and the CIFAR-10 dataset for target classification is used to evaluate the performance.Experimental results confirm the convergence and the accuracy of the model using the pseudo-label based semi-supervised learning in the DML framework.Given the proportion of the pseudo-label dataset is 20%,the accuracy of the model is over 90% when the value of local parameter update steps between two global aggregations is less than 5.Besides,fixing the global aggregations interval to 3,the model converges with acceptable performance degradation when the proportion of the pseudo-label dataset varies from 20% to 80%.

Detecting and Mitigating DDOS Attacks in SDNs Using Deep Neural Network

Distributed denial of service(DDoS)attack is the most common attack that obstructs a network and makes it unavailable for a legitimate user.We proposed a deep neural network(DNN)model for the detection of DDoS attacks in the Software-Defined Networking(SDN)paradigm.SDN centralizes the control plane and separates it from the data plane.It simplifies a network and eliminates vendor specification of a device.Because of this open nature and centralized control,SDN can easily become a victim of DDoS attacks.We proposed a supervised Developed Deep Neural Network(DDNN)model that can classify the DDoS attack traffic and legitimate traffic.Our Developed Deep Neural Network(DDNN)model takes a large number of feature values as compared to previously proposed Machine Learning(ML)models.The proposed DNN model scans the data to find the correlated features and delivers high-quality results.The model enhances the security of SDN and has better accuracy as compared to previously proposed models.We choose the latest state-of-the-art dataset which consists of many novel attacks and overcomes all the shortcomings and limitations of the existing datasets.Our model results in a high accuracy rate of 99.76%with a low false-positive rate and 0.065%low loss rate.The accuracy increases to 99.80%as we increase the number of epochs to 100 rounds.Our proposed model classifies anomalous and normal traffic more accurately as compared to the previously proposed models.It can handle a huge amount of structured and unstructured data and can easily solve complex problems.

结合概率图模型与DNN的DDoS攻击检测方法

从传统网络到物联网,分布式拒绝服务攻击一直是网络安全的隐患。为提高分布式拒绝服务攻击的检测率,提出基于概率图模型与深度神经网络的DDoS攻击检测方案。该检测方案由数据预处理阶段和攻击检测阶段组成,在数据预处理阶段,研究了正常数据包与攻击包的区别,分别从TCP、UDP以及IP数据包包头信息提取出较高维的统计特征,根据随机森林计算的特征重要性因子,保留了前22个特征用于流量检测。22个统计特征通过概率图模型的隐马尔科夫算法进行聚类,然后将聚类结果通过检测阶段的深度神经网络对网络数据进行进一步的检测。在CICDoS数据集上进行验证性实验,结果表明,该检测方法的准确率最高可达99.35%,最低检测误报率和漏警率分别可达0.51%和0.12%。

生物氧化预处理过程pH值随机分布控制方法研究

生物氧化预处理过程中氧化槽pH值是影响细菌活性的关键因素之一,而pH值输出形态分布不符合高斯分布,使传统的均值和方差难以描述输出pH值分布,本文提出一种对矿浆输出pH的概率密度函数(PDF)统计信息控制方法。首先,采用B样条逼近矿浆输出pH值的PDF统计信息;其次,针对权值向量之间的关系,利用动态神经网络(DNN)建立控制输入和权值向量之间的非线性动态模型,基于建立pH的PDF统计信息权值模型,设计滑模变结构控制器,通过构造Lyapunov函数进行稳定性分析;最后,实现输出PDF统计信息对目标PDF统计信息的跟踪。仿真结果验证了所提方法的有效性,为生物氧化预处理过程提供了新方法。

Distributed Heterogeneous Multi-Agent Optimization with Stochastic Sub-Gradient

This paper studies the optimization problem of heterogeneous networks under a timevarying topology.Each agent only accesses to one local objective function,which is nonsmooth.An improved algorithm with noisy measurement of local objective functions' sub-gradients and additive noises among information exchanging between each pair of agents is designed to minimize the sum of objective functions of all agents.To weaken the effect of these noises,two step sizes are introduced in the control protocol.By graph theory,stochastic analysis and martingale convergence theory,it is proved that if the sub-gradients are uniformly bounded,the sequence of digraphs is balanced and the union graph of all digraphs is joint strongly connected,then the designed control protocol can force all agents to find the global optimal point almost surely.At last,the authors give some numerical examples to verify the effectiveness of the stochastic sub-gradient algorithms.

针对样本类不平衡的深度残差网络电力系统暂态稳定评估方法

系统的量测数据可能受到噪声以及样本类分布不平衡问题的影响,导致基于数据驱动的暂态稳定评估模型性能下降。提出一种针对样本类不平衡的的深度残差网络电力系统暂态稳定评估方法。首先,利用改进过采样技术为滤除噪声的少数类样本构造所需的新样本,改善样本类不平衡问题,并减少噪声的影响;然后,基于深度残差网络构建电力系统暂态稳定评估模型,解决梯度消失导致的模型性能退化问题,提高模型的鲁棒性和准确性;最后,在新英格兰10机39节点和47机140节点系统上的仿真结果表明,所提方法能减小噪声干扰、降低不平衡数据集所带来的影响和减少计算复杂度。

SDN中基于信息熵与DNN的DDoS攻击检测模型

软件定义网络(software defined networking, SDN)解耦了网络的数据层与控制层,同时控制器也面临"单点失效"的危险.攻击者可以发起分布式拒绝服务攻击(distributed denial of service, DDoS)使控制器失效,影响网络安全.为解决SDN中的DDoS流量检测问题,创新性地提出了基于信息熵与深度神经网络(deep neural network, DNN)的DDoS检测模型.该模型包括基于信息熵的初检模块和基于深度神经网络DNN的DDoS流量检测模块.初检模块通过计算数据包源、目的IP地址的信息熵值初步发现网络中的可疑流量,并利用基于DNN的DDoS检测模块对疑似异常流量进行进一步确认,从而发现DDoS攻击.实验表明:该模型对DDoS流量的识别率达到99%以上,准确率也有显著提高,误报率明显优于基于信息熵的检测方法.同时,该模型还能缩短检测时间,提高资源使用效率.

基于深度学习的配电网安全态势感知研究

配电网作为电力系统的关键环节,有必要识别配电网潜在危害,避免失稳。为了解决数据中噪声干扰的问题并提高态势预测准确性,提出了一种基于深度学习的配电网安全态势感知方法。首先,采集配电网运行量,利用奇异值分解(singular value decomposition,SVD)对运行量进行降噪;其次,分析运行量与安全态势的关系,采用评估值指标评估配电网态势;最后,利用注意力时域卷积网络(temporal convolution network-attention mechanism,TCNAM)对降噪后的输入数据预测得出态势评估值,预测配电网潜在危害,若失稳,则发出预警信号。通过对IEEE 33节点系统和实际配电网系统仿真可知,TCN-AM预测效果好,且进行降噪处理后预测准确性有所提高,能够在满足预警条件后,发出相应的预警信号。所提方法在降噪处理后能够更准确地实现配电网的安全态势感知。

Distribution-Transformed Network for Impulse Noise Removal

This work aims to explore the restoration of images corrupted by impulse noise via distribution-transformed network (DTN), which utilizes convolutional neural network to learn pixel-distribution features from noisy images. Compared with the traditional median-based algorithms, it avoids the complicated pre-processing procedure and directly tackles the original image. Additionally, different from the traditional methods utilizing the spatial neighbor information around the pixels or patches and optimizing in an iterative manner, this work turns to capture the pixel-level distribution information by means of wide and transformed network learning. DTN fits the distribution at pixel-level with larger receptions and more channels. Furthermore, DTN utilities a residual block without batch normalization layer to generate a good estimate. In terms of edge preservation and noise suppression, the proposed DTN consistently achieves significantly superior performance than current state-of-the-art methods, particularly at extreme noise densities.

基于SINR的动态无线网络分布式链路调度

无线信号之间的干扰阻碍了信号的并发传输,降低了无线网络的吞吐量.链路调度是提高无线网络吞吐量、减少信号传输延迟的一种有效方法.因为SINR (signal to interference plus noise ratio)模型准确地描述了无线信号传播的固有特性,能够真实反映无线信号之间的干扰,提出一种在动态无线网络中基于SINR模型的常数近似因子的在线分布式链路调度算法(OLD_LS).在线的意思是指,在算法执行的过程中任意节点可以随时加入网络,也可以随时离开网络.节点任意加入网络或者从网络中离开体现了无线网络的动态变化的特性. OLD_LS算法把网络区域划分为多个正六边形,局部化SINR模型的全局干扰.设计动态网络下的领导者选举算法(LE),只要网络节点的动态变化速率小于1/ε, LE就可以在O(log n+log R)ε≤5(1-21-α/2)/6,α表示路径损耗指数, n是网络节点的规模, R是最长链路的长度.根据文献调研,所提算法是第1个用于动态无线网络的在线分布式链路调度算法.

冲击噪声下基于演化长短时记忆神经网络的调制信号识别

为了解决冲击噪声下长短时记忆(long short term memory,LSTM)神经网络调制信号识别方法抗冲击噪声能力弱和超参数难以确定的问题,本文提出了一种演化长短时记忆神经网络的调制识别方法。利用基于短时傅里叶变换的卷积神经网络(convolution neural network,CNN)去噪模型对数据集去噪;结合量子计算机制和旗鱼优化器(sailfish optimizer,SFO)设计了量子旗鱼算法(quantum sailfish algorithm,QSFA)去演化LSTM神经网络以获得最优的超参数;使用演化长短时记忆神经网络作为分类器进行自动调制信号识别。仿真结果表明,采用所设计的CNN去噪和演化长短时记忆神经网络模型,识别准确率有了大幅度的提高。量子旗鱼算法演化LSTM神经网络模型降低了传统LSTM神经网络容易陷于局部极小值或者过拟合的概率,当混合信噪比为0 dB,所提方法对11种调制信号的平均识别准确率达到90%以上。

基于t-SSA-BP的煤矿噪声职业健康损害预测

为精准预测煤矿接噪人员职业健康损害情况,基于分位图法选取健康损害数据影响因素,依据噪声职业健康损害评估方法,以自适应t分布变异因子和麻雀搜索算法(SSA)作为参数优化算法,建立t-SSA-BP煤矿噪声职业健康损害预测模型,利用基准函数测试算法寻优性能,并以陕北地区10个煤矿为研究对象,采用现场调查、理论分析和Matlab仿真模拟方法验证煤矿噪声职业健康损害预测模型。结果表明:噪声暴露强度、个体年龄、接噪工龄和接噪岗位4个指标是煤矿噪声健康损害的影响因素;t-SSA较SSA在4种基准函数上整体精度提升660%,5种噪声健康损害神经网络预测模型预测精度从高到低依次为:t-SSA-BP>SSA-BP>PSO-BP>CFA-PSO-RBF>PSO-GRNN,t-SSA-BP预测模型的平均绝对误差(MAE)、平均绝对百分误差(MAPE)相比SSA-BP分别降低681%、667%,决定系数(R2)达0999,预测精度明显提升,且收敛速度更快。

基于多尺度时频图与卷积神经网络的车轮故障智能诊断

铁道车辆车轮故障的产生,不仅会增大列车的振动和噪声使乘坐舒适性下降,而且会加速车辆及轨道零部件的损伤,严重时还会引发事故,因此对车轮服役状态的实时监测对保证列车安全运营具有重要意义。针对现有铁道车辆车轮故障诊断方法存在自适应能力弱、准确率低等不足,提出一种基于多尺度时频图与卷积神经网络(CNN)相结合的车轮故障智能诊断方法,该方法利用车轮所在轴箱垂向振动加速度来间接识别车轮服役状态。1)首先采用形态学滤波器对车辆轴箱振动加速度信号进行滤波降噪,然后采用完全噪声辅助聚合经验模态分解(CEEMDAN)将滤波后的信号自适应地分解为若干固有模态函数(IMF),选取能量熵增量相对较大的三阶分量作为信号的主分量。2)分别求各主分量的Wigner-Ville分布(WVD),然后叠加转化为多尺度时频图。3)对经典的LeNet-5模型进行结构改进和网络参数优化,构建适合车轮故障诊断的CNN模型,来学习提取车轮在不同工况下的时频图特征,并对时频图进行分类,将特征学习提取与故障分类融为一体,一定程度上实现了端到端的车轮故障诊断。经仿真试验和现场试验验证表明:所提出的方法对于车速、故障类型和故障程度都有很好的自适应能力,故障识别准确率可达97%,且泛化能力强。因此所提方法在车辆运营状态在线监测的应用中具有一定的理论意义和工程价值。

一种融合噪声网络的深度强化学习通信干扰资源分配算法

针对传统干扰资源分配算法在处理非线性组合优化问题时需要较完备的先验信息,同时决策维度小,无法满足现代通信对抗要求的问题,该文提出一种融合噪声网络的深度强化学习通信干扰资源分配算法(FNNDRL)。借鉴噪声网络的思想,该算法设计了孪生噪声评估网络,在避免Q值高估的基础上,通过提升评估网络的随机性,保证了训练过程的探索性;基于概率熵的物理意义,设计了基于策略分布熵改进的策略网络损失函数,在最大化累计奖励的同时最大化策略分布熵,避免策略优化过程中收敛到局部最优。仿真结果表明,该算法在解决干扰资源分配问题时优于所对比的平均分配和强化学习方法,同时算法稳定性较高,对高维决策空间适应性强。

基于生成对抗网络与噪声分布的图像超分辨率重建方法

现有的图像超分辨率重建方法都较少考虑真实低分辨率图像中包含的噪声信息,因此会影响图像的重建质量.受真实图像去噪算法的启发,本文引入一个噪声分布收集网络来收集低分辨率图像的噪声分布信息,并采用生成对抗网络的模型设计,提高含噪声图像的重建质量.噪声分布信息会分别输入到超分辨率重建网络和判别网络,在重建过程中去除噪声的同时保证有用高频信息的恢复,另外由于判别网络的能力对整个模型的性能有着重要影响,选择使用U-Net网络来获得更好的梯度信息反馈.与经典图像超分辨率重建算法的对比以及消融实验表明,使用噪声收集网络和U-Net判别网络后,本文模型在噪声低分辨率图像重建任务中获得了更好的性能.

基于混合神经网络的电力线脉冲噪声参数估计方法

脉冲噪声广泛存在于电力线通信(power line communication,PLC)系统中,会严重影响系统的通信性能。电力线脉冲噪声的建模通常使用α稳定分布模型,为达到最佳的脉冲噪声抑制效果,需要知道脉冲噪声的类型和相关参数。为此,文章提出一种基于混合神经网络的符合α稳定分布的脉冲噪声参数估计方法。不同于传统的方法,本方法可以分别独立地估计α稳定分布的重要参数α(即特征指数)和γ(即尺度参数)。仿真结果表明,与传统方法相比,提出的方法具有更准确的参数估计性能,归一化均方误差值仅为10–4左右。

Novel Affine Power Flow Method for Improving Accuracy of Interval Power Flow Data in Cyber Physical Systems of Active Distribution Networks

A large number of load power and power output of distributed generation in an active distribution network(ADN)are uncertain,which causes the classical affine power flow method to encounter problems of interval expansion and low efficiency when applied to an AND.This then leads to errors of interval power flow data sources in the cyber physical system(CPS)of an ADN.In order to improve the accuracy of interval power flow data in the CPS of an ADN,an affine power flow method of an ADN for restraining interval expansion is proposed.Aiming at the expansion of interval results caused by the approximation error of non-affine operations in an affine power flow method,the approximation method of the new noise source coefficient is improved,and it is proved that the improved method is superior to the classical method in restraining interval expansion.To overcome the decrease of computational efficiency caused by new noise sources,a novel merging method of new noise sources in an iterative process is designed.Simulation tests are conducted on an IEEE 33-bus,PG&E 69-bus and an actual 1180-bus system,which proves the validity of the proposed affine power flow method and its advantages in terms of computational efficiency and restraining interval expansion.

抑制用电负载噪声的低压断路器装置设计与应用

针对低压台区电力线载波信道易受居民用电负载接入噪声干扰,导致部分台区难以满足新型电力系统“源-网-荷”高效互动所需的实时载波通信需求的问题,提出一种抑制用电负载噪声的断路器装置设计与应用方法。首先对用电负载接入噪声特性及其对载波通信影响进行了分析;然后设计了抑制用电负载噪声的断路器装置,对断路器装置电路的原理和参数进行了详细设计,该装置在电表外置断路器基础上增加了低通滤波、阻抗匹配、载波信号旁路电路,在抑制居民用电负载噪声的同时构建连通户内电器和电表之间的载波信号旁路通道;最后结合典型案例台区对文中断路器装置的使用效果进行了分析。应用结果表明,该断路器装置可有效提升台区电力线载波通信的成功率和网络性能,减少现场维护工作量,为建设新型电力系统的高效本地通信网络提供技术支持。

最小错误准则下多传感器信号检测与调制识别

针对多传感器分布式接收中的弱信号检测与调制识别稳健性不足的问题,提出了一种基于深度学习的联合处理算法。该方法采用分布式软信息融合处理策略,将信号检测与调制识别综合为一个多元假设检验问题,借助深度神经网络优异的函数逼近能力,在对网络结构、目标函数和网络输入输出进行分析基础上,给出了基于深度神经网络的联合后验概率求解及分类判决方法。通过仿真实验对所提方法性能进行了验证,结果表明,该方法能实现多个传感器信号有效融合,并且随着接收单元数目增加,分类准确率明显提升;与现有基于等权值合并的置信度融合方法相比,该方法性能更优,且在低信噪比、短数据和接收单元数目较多时优势体现更加明显。