Stability of Iterative Learning Control with Data Dropouts via Asynchronous Dynamical System

In this paper, the stability of iterative learning control with data dropouts is discussed. By the super vector formulation, an iterative learning control （ILC） system with data dropouts can be modeled as an asynchronous dynamical system with rate constraints on events in the iteration domain. The stability condition is provided in the form of linear matrix inequalities （LMIS） depending on the stability of asynchronous dynamical systems. The analysis is supported by simulations.

Optimal full-order filtering for discrete-time systems with random measurement delays and multiple packet dropouts

This paper is concerned with the estimation problem for discrete-time stochastic linear systems with possible single unit delay and multiple packet dropouts. Based on a proposed uncertain model in data transmission, an optimal full-order filter for the state of the system is presented, which is shown to be of the form of employing the received outputs at the current and last time instants. The solution to the optimal filter is given in terms of a Riccati difference equation governed by two binary random variables. The optimal filter is reduced to the standard Kalman filter when there are no random delays and packet dropouts. The steady-state filter is also investigated. A sufficient condition for the existence of the steady-state filter is given. The asymptotic stability of the optimal filter is analyzed.

Fault detection for networked systems subject to access constraints and packet dropouts

This paper addresses the problem of fault detection（FD） for networked systems with access constraints and packet dropouts.Two independent Markov chains are used to describe the sequences of channels which are available for communication at an instant and the packet dropout process,respectively.Performance indexes H∞ and H_ are introduced to describe the robustness of residual against external disturbances and sensitivity of residual to faults,respectively.By using a mode-dependent fault detection filter（FDF） as residual generator,the addressed FD problem is converted into an auxiliary filter design problem with the above index constraints.A sufficient condition for the existence of the FDF is derived in terms of certain linear matrix inequalities（LMIs）.When these LMIs are feasible,the explicit expression of the desired FDF can also be characterized.A numerical example is exploited to show the usefulness of the proposed results.

On the loss mechanisms of radiation belt electron dropouts during the 12 September 2014 geomagnetic storm

Radiation belt electron dropouts indicate electron flux decay to the background level during geomagnetic storms,which is commonly attributed to the effects of wave-induced pitch angle scattering and magnetopause shadowing.To investigate the loss mechanisms of radiation belt electron dropouts triggered by a solar wind dynamic pressure pulse event on 12 September 2014,we comprehensively analyzed the particle and wave measurements from Van Allen Probes.The dropout event was divided into three periods:before the storm,the initial phase of the storm,and the main phase of the storm.The electron pitch angle distributions(PADs)and electron flux dropouts during the initial and main phases of this storm were investigated,and the evolution of the radial profile of electron phase space density(PSD)and the(μ,K)dependence of electron PSD dropouts(whereμ,K,and L^*are the three adiabatic invariants)were analyzed.The energy-independent decay of electrons at L>4.5 was accompanied by butterfly PADs,suggesting that the magnetopause shadowing process may be the major loss mechanism during the initial phase of the storm at L>4.5.The features of electron dropouts and 90°-peaked PADs were observed only for>1 MeV electrons at L<4,indicating that the wave-induced scattering effect may dominate the electron loss processes at the lower L-shell during the main phase of the storm.Evaluations of the(μ,K)dependence of electron PSD drops and calculations of the minimum electron resonant energies of H+-band electromagnetic ion cyclotron(EMIC)waves support the scenario that the observed PSD drop peaks around L^*=3.9 may be caused mainly by the scattering of EMIC waves,whereas the drop peaks around L^*=4.6 may result from a combination of EMIC wave scattering and outward radial diffusion.

Observer-based H-infinity control in multiple channel networked control systems with random packet dropouts

This paper investigates the observer-based H-infinity control problem for networked control systems （NCSs） with random packet dropouts. A general packet dropout model with multiple independent stochastic variables in the multiple channels case is adopted to describe the data missing in the limited communication channels. With the consideration of the sensor-to-controller and controller-to-actuator packet dropouts at the same time, a new method is pro- posed based on a separation lemma to design an observer-based H-infinity controller, which exponentially stabilizes the closed-loop system in the sense of mean square and also achieves a prescribed H-infinity disturbance attenuation level. A numerical example is given to illustrate the effectiveness of the proposed control method.

Chinese Students in Japan Help School Dropouts at Home

In 2004, Wang Chengyan, a 13-year-old Mongolian girl in the Tumote Left Banner of Inner Mongolia, took up her schoolbag again and marched into the classroom of six grade of a local primary school. With her face shining with brilliance, she told her friends： ＂It is brothers and sisters studying in Japan who have paid my way to school.＂

Impulsive controller design for nonlinear networked control systems with time delay and packet dropouts

The globally exponential stability of nonlinear impul- sive networked control systems （NINCS） with time delay and packet dropouts is investigated. By applying Lyapunov function theory, sufficient conditions on the global exponential stability are derived by introducing a comparison system and estimating the corresponding Cauchy matrix. An impulsive controller is explicitly designed to achieve exponential stability and ensure state con- verge with a given decay rate for the system. The Lorenz oscillator system is presented as a numerical example to illustrate the theo- retical results and effectiveness of the proposed controller design procedure.

Enabling Proactive Management of School Dropouts Using Neural Network

The growing need to use Artificial Intelligence (AI) technologies in addressing challenges in education sectors of developing countries is undermined by low awareness, limited skill and poor data quality. One particular persisting challenge, which can be addressed by AI, is school dropouts whereby hundreds of thousands of children drop annually in Africa. This article presents a data-driven approach to proactively predict likelihood of dropping from schools and enable effective management of dropouts. The approach is guided by a carefully crafted conceptual framework and new concepts of average absenteeism, current cumulative absenteeism and dropout risk appetite. In this study, a typical scenario of missing quality data is considered and for which synthetic data is generated to enable development of a functioning prediction model using neural network. The results show that, using the proposed approach, the levels of risk of dropping out of schools can be practically determined using data that is largely available in schools. Potentially, the study will inspire further research, encourage deployment of the technologies in real life, and inform processes of formulating or improving policies.

基于卷积神经网络的特定目标文本情感分析模型

在特定目标文本情感分析过程中,文本序列分类受标注方式的限制,导致分析结果的准确率和召回率较低.为了解决这个问题,构建了基于卷积神经网络的特定目标文本情感分析模型(文本分析模型).根据情感差异分析特定目标文本序列,在输入层将文本特征矩阵作为卷积神经网络语言模型的输入数据,拼接成词性序列矩阵;分段池化捕获文本序列不同的关键特征,并分类处理提取到的特征向量;加入dropout机制完成特定目标文本情感分类,确定文本中每个词的重要度信息,实现特定目标文本情感分析.实验结果表明,文本分析模型的准确率高于84%,召回率最大值为87%,能够有效实现特定目标文本情感分析.

GPU异构计算环境中长短时记忆网络模型的应用及优化

随着深度学习的广泛应用及算力资源的异构化,在GPU异构计算环境下的深度学习加速成为又一研究热点。文章探讨了在GPU异构计算环境中如何应用长短时记忆网络模型,并通过优化策略提高其性能。首先,介绍了长短时记忆网络模型的基本结构(包括门控循环单元、丢弃法、Adam与双向长短时记忆网络等);其次,提出了在GPU上执行的一系列优化方法,如CuDNN库的应用及并行计算的设计等。最终,通过实验分析了以上优化方法在训练时间、验证集性能、测试集性能、超参数和硬件资源使用等方面的差异。

基于梯度选择的轻量化差分隐私保护联邦学习

为了应对机器学习过程中可能出现的用户隐私问题,联邦学习作为首个无需用户上传真实数据、仅上传模型更新的协作式在线学习解决方案,已经受到人们的广泛关注与研究。然而,它要求用户在本地训练且上传的模型更新中仍可能包含敏感信息,从而带来了新的隐私保护问题。与此同时,必须在用户本地进行完整训练的特点也使得联邦学习过程中的运算与通信开销问题成为一项挑战,亟需人们建立一种轻量化的联邦学习架构体系。出于进一步的隐私需求考虑,文中使用了带有差分隐私机制的联邦学习框架。另外,首次提出了基于Fisher信息矩阵的Dropout机制——FisherDropout,用于对联邦学习过程中在客户端训练产生梯度更新的每个维度进行优化选择,从而极大地节约运算成本、通信成本以及隐私预算,建立了一种兼具隐私性与轻量化优势的联邦学习框架。在真实世界数据集上的大量实验验证了该方案的有效性。实验结果表明,相比其他联邦学习框架,FisherDropout机制在最好的情况下可以节约76.8%~83.6%的通信开销以及23.0%~26.2%的运算开销,在差分隐私保护中隐私性与可用性的均衡方面同样具有突出优势。

Fault Detection for Uncertain Delta Operator Systems with Two-Channel Packet Dropouts via a Switched Systems Approach

This paper utilizes a switched systems approach to deal with the problem of fault detectio for uncertain delta operator networked control system with packet dropouts and timevarying delays.Uncertainties exist in the matrices of the systems and are norm-bounded time-varying.Two parts of packet dropouts are considered in this paper:From sensors to controllers,and from controllers to actuators.Two independent Bernoulli distributed white sequences are introduced to account for packet dropouts.Then an FD filter is designed under an arbitrary switching law.Furthermore,the sufficient conditions for the NCSs under consideration that are exponentially stable in the mean-square sense and satisfy H∞performance are obtained in terms of linear matrix inequalitie,multiple Lyapunov function and average dwell-tim approach.The explicit expression of the desired filter parameters is given.Finally,a numerical example verifies the effectiveness of the proposed method.

基于添加Dropout层的CNN-LSTM网络短期负荷预测

精准的短期负荷预测能帮助电力部门制订合理的生产调度计划,达到节省能源消耗的目的。为提升短期负荷预测的准确性,提出一种添加Dropout层的CNN-LSTM网络短期负荷预测方法。首先,根据电力负荷预测流程对预测的影响因素如气象、日期类型等进行相关性验证后构建输入特征向量;其次,使用一维卷积网络对输入的特征向量进行卷积、池化处理,并在LSTM网络全连接层添加Dropout层进行短期负荷预测仿真实验;最后,使用某电网历史数据进行测试。结果表明,相比于单独的LSTM网络,所建模型对短期负荷的预测效果更好。

One memristor–one electrolyte-gated transistor-based high energy-efficient dropout neuronal units

Artificial neural networks(ANN) have been extensively researched due to their significant energy-saving benefits.Hardware implementations of ANN with dropout function would be able to avoid the overfitting problem. This letter reports a dropout neuronal unit(1R1T-DNU) based on one memristor–one electrolyte-gated transistor with an ultralow energy consumption of 25 p J/spike. A dropout neural network is constructed based on such a device and has been verified by MNIST dataset, demonstrating high recognition accuracies(> 90%) within a large range of dropout probabilities up to40%. The running time can be reduced by increasing dropout probability without a significant loss in accuracy. Our results indicate the great potential of introducing such 1R1T-DNUs in full-hardware neural networks to enhance energy efficiency and to solve the overfitting problem.

基于CNN与Bi-LSTM的异常用电行为甄别算法研究

为降低非技术损耗给电网运行带来的损失,文中对用户用电行为的甄别技术进行了研究,并提出了一种基于卷积神经网络(CNN)和双向长短期记忆(Bi-LSTM)网络的模型。该模型一方面使用CNN中的卷积、池化运算提升对用电数据中隐性特征的挖掘效率,另一方面利用Bi-LSTM处理长时间序列的优势,弥补了CNN在时序分析上的不足。同时还引入了Dropout机制与Adam优化方法,提升了网络的训练速度,避免了CNN和Bi-LSTM结合后因网络结构复杂而造成的过拟合现象。在自建数据集上进行的仿真结果表明,所提算法的Precision、Recall及F1值均取得了显著提升,且相较于单一的CNN和Bi-LSTM网络,F1值分别提升了10.04%和8.32%。此外,该算法训练与测试的F1值仅相差0.03%,说明算法的鲁棒性较强,未出现过拟合的现象。

基于改进降噪自编码器的馈线终端失效率预测

配电网中馈线终端设备由于运行环境恶劣,往往面临意外失效问题。本文针对海量馈线终端装置的失效率预测问题,使用堆叠降噪自编码器实现基于馈线终端的各个关键元件的失效率预测;采用基于Dropout的模型正则化方法防止自编码器训练过程中出现过拟合现象,同时采用Adadelta算法对堆叠自编码器进行优化,在保证预测准确率的同时提高学习速率,实现馈线终端故障失效率的高效准确预测;最后基于馈线终端装置现场数据进行仿真验证。仿真结果验证了本文方法对失效率预测的准确性和泛化能力。

应用改进卷积神经网络的客户服务业务中台资源异常信息主动报警

针对客户服务业务中台资源异常信息人工诊断不及时、故障辨识率低等问题,提出一种基于改进卷积神经网络的故障诊断方法。卷积层后引入批量归一化层提高模型的泛化能力,在全连接层引入Droupout函数来缓解过拟合问题,还对数据进行了增强处理以及运用灰狼算法对超参数进行寻优。该模型在Pytorch和Pycharm环境下进行仿真,得出经典卷积神经网络的测试集准确率在85%左右,而改进后的测试集准确率在94%左右,表明所提设计具有明显效果。

Optimal fusion state estimator for a multi-sensor system subject to multiple packet dropouts

In this note,we study the state estimation problem for a multi-sensor system subject to multiple packet dropouts.A novel optimal distributed fusion estimator is derived by applying a resending mechanism and a parallel information filtering structure.It is shown that the proposed distributed fusion estimator has smaller estimation error covariance and less computation complexity when compared with the centralised Kalman like estimator with multiple intermittent measurements.

Deep Learning Recognition for Arabic Alphabet Sign Language RGB Dataset

This paper introduces a Convolutional Neural Network (CNN) model for Arabic Sign Language (AASL) recognition, using the AASL dataset. Recognizing the fundamental importance of communication for the hearing-impaired, especially within the Arabic-speaking deaf community, the study emphasizes the critical role of sign language recognition systems. The proposed methodology achieves outstanding accuracy, with the CNN model reaching 99.9% accuracy on the training set and a validation accuracy of 97.4%. This study not only establishes a high-accuracy AASL recognition model but also provides insights into effective dropout strategies. The achieved high accuracy rates position the proposed model as a significant advancement in the field, holding promise for improved communication accessibility for the Arabic-speaking deaf community.

基于Bi-LSTM-Dropout的蓄电池剩余使用寿命预测方法

蓄电池剩余使用寿命预测对能源的安全性和可持续发展至关重要。该文提出一种蓄电池剩余使用寿命的预测方法,利用蓄电池的历史运行数据和充放电周期,构建Bi-LSTM-Dropout网络模型。利用Bi-LSTM提取时间序列中蓄电池长期依赖的特征,采用Dropout优化算法降低Bi-LSTM网络模型的复杂度,提高模型的泛化能力。实验结果表明,该方法在测试集上的准确率达96.2%,实现了蓄电池剩余使用寿命的精确预测。