Fully Distributed Learning for Deep Random Vector Functional-Link Networks

In the contemporary era, the proliferation of information technology has led to an unprecedented surge in data generation, with this data being dispersed across a multitude of mobile devices. Facing these situations and the training of deep learning model that needs great computing power support, the distributed algorithm that can carry out multi-party joint modeling has attracted everyone’s attention. The distributed training mode relieves the huge pressure of centralized model on computer computing power and communication. However, most distributed algorithms currently work in a master-slave mode, often including a central server for coordination, which to some extent will cause communication pressure, data leakage, privacy violations and other issues. To solve these problems, a decentralized fully distributed algorithm based on deep random weight neural network is proposed. The algorithm decomposes the original objective function into several sub-problems under consistency constraints, combines the decentralized average consensus (DAC) and alternating direction method of multipliers (ADMM), and achieves the goal of joint modeling and training through local calculation and communication of each node. Finally, we compare the proposed decentralized algorithm with several centralized deep neural networks with random weights, and experimental results demonstrate the effectiveness of the proposed algorithm.

Linearization Learning Method of BP Neural Networks

Feedforward multi layer neural networks have very strong mapping capability that is based on the non linearity of the activation function, however, the non linearity of the activation function can cause the multiple local minima on the learning error surfaces, which affect the learning rate and solving optimal weights. This paper proposes a learning method linearizing non linearity of the activation function and discusses its merits and demerits theoretically.

Study on Decision Method of Neutral Point Grounding Mode for Medium-Voltage Distribution Network

The neutral grounding mode of medium-voltage distribution network decides the reliability, overvoltage, relay protection and electrical safety. Therefore, a comprehensive consideration of the reliability, safety and economy is particularly important for the decision of neutral grounding mode. This paper proposes a new decision method of neutral point grounding mode for mediumvoltage distribution network. The objective function is constructed for the decision according the life cycle cost. The reliability of the neutral point grounding mode is taken into account through treating the outage cost as an operating cost. The safety condition of the neutral point grounding mode is preserved as the constraint condition of decision models, so the decision method can generate the most economical and reliable scheme of neutral point grounding mode within a safe limit. The example is used to verify the feasibility and effectiveness of the decision method.

Calculation method of ship collision force on bridge using artificial neural network

Ship collision on bridge is a dynamic process featured by high nonlinearity and instantaneity. Calculating ship-bridge collision force typically involves either the use of design-specification-stipulated equivalent static load, or the use of finite element method (FEM) which is more time-consuming and requires supercomputing resources. In this paper, we proposed an alternative approach that combines FEM with artificial neural network (ANN). The radial basis function neural network (RBFNN) employed for calculating the impact force in consideration of ship-bridge collision mechanics. With ship velocity and mass as the input vectors and ship collision force as the output vector, the neural networks for different network parameters are trained by the learning samples obtained from finite element simulation results. The error analyses of the learning and testing samples show that the proposed RBFNN is accurate enough to calculate ship-bridge collision force. The input-output relationship obtained by the RBFNN is essentially consistent with the typical empirical formulae. Finally, a special toolbox is developed for calculation efficiency in application using MATLAB software.

一种面向SDN的传统网络功能集成方法设计与实现

软件定义网络(SDN)通过将控制平面与数据平面分离并使网络可编程来改变网络的管理方式。控制器是SDN网络的关键组成部分。对于诸如DHCP中继功能、动态路由功能等传统网络功能,目前已存在成熟的第三方网络组件。基于能够更便捷地开发面向SDN的传统网络应用,设计与实现一种通用的传统网络功能集成方法,该方法通过将底层网络流量同步给第三方网络组件,由第三方网络组件完成核心的网络功能。文中基于ONOS控制器利用该集成方法开发了一款通用的传统网络应用,并通过集成DHCP Relay功能的实验完成了对该集成方法的功能性验证。

不确定电液伺服系统的时变输出约束自适应滤波控制

针对电液伺服系统位置跟踪控制中存在的输出约束和不确定性问题,提出一种基于正切型时变障碍Lyapunov函数的输出约束自适应滤波控制方法。构造具有时变约束边界的正切型时变障碍Lyapunov函数,通过时变边界函数的参数设置,使系统输出具有较好的瞬态和稳态性能;设计径向基函数(RBF)神经网络及权重自适应学习律,在线逼近由模型不确定性和未知干扰组成的复合干扰,并将逼近值用于反馈控制;采用二阶指令滤波反步法设计状态反馈控制律和误差补偿机制,避免反步设计中“计算爆炸”的问题,同时消除滤波误差,提高系统位置跟踪精度;依据Lyapunov稳定性理论证明闭环系统中所有误差信号的收敛性。仿真结果表明:系统的稳态误差在所提方法下约为3.48×10^(-8)m,相比于其他控制方法,跟踪误差始终约束在时变的约束边界内,跟踪精度和控制性能均得到提升。

基于改进残差网络的运动目标模糊图像复原方法

传统的残差网络在复原运动目标模糊图像时,在模糊程度较严重的情况下,存在特征提取不充分、噪声干扰等问题,导致恢复出的图像无法完全达到原始图像的清晰度和细节。对此,提出基于改进残差网络的运动目标模糊图像复原方法。对采集到的运动目标模糊图像,采用多损失函数融合方法改进传统残差块结构,构建编码器-解码器网络训练结构,训练损失函数,提升网络的特征学习能力。通过完成训练的网络,输出运动目标模糊图像复原结果。实验结果表明,该方法复原运动目标模糊图像的峰值信噪比高于30 dB,结构相似性高于0.9。

反向传播神经网络的太平洋海域温跃层反演

温跃层是反映海洋温度场的重要指标。针对太平洋中东部海域复杂多变的水文情况以及传统温跃层分析模式的局限性,该文基于BOA_Argo历史网格,通过BP神经网络,建立温度剖面的经验正交系数与海面遥感数据、少量深度处海水温度之间的非线性映射关系,实现海洋垂向温度剖面的实时反演,最后利用垂向梯度法获得海洋温跃层的相关参数。实验结果表明,相比于传统方法,该方法反演得到的跃层深度与测量值更加吻合,其中上层深度平均反演误差从10.3 m下降到5.7 m,下层深度平均反演误差从16.8 m下降到8.8 m。

面向深度神经网络的电力芯片功能检测方法

电力芯片功能检测对于保证电力正常运行有重要意义,目前的方法检测准确率相对较低,导致检测时间过长。为了解决上述问题,面向深度神经网络提出了一种新的电力芯片功能检测方法,计算电力芯片功能指标参数,确定电力芯片功能检测的真实数据,将独立神经元进行离散运算,并对其粒子量进行计算。根据得到的数据信息分析检测的隐藏神经元,计算回路的品质因数或谐振系数,检测传输功能,根据得到的实际处理量检测电力芯片的处理功能,通过建立待测量曲线和日负荷数据曲线,确定异常数据筛选功能。实验结果表明,所设计方法传输功能和处理功能检测准确率在98%以上,异常数据筛选功能检测准确率在99%以上,当检测数据量大于300 GB时,检测时间低于0.5s,所研究方法具有较好的性能。

基于改进深度动态模糊神经网络的信息综合分析算法

针对传统人力资源评价算法的主观性强,难以反映员工真实能力的问题,提出了一种结合深度动态模糊神经网络和粒子群优化的算法。该算法对传统模糊神经网络进行了改进,并使用动态结构来增强原模型训练能力,通过对隶属函数层的优化,使模型具备了处理广域数据的能力。为了提高算法的运行效率,还采用误差下降法对模型的规则权重进行排序并完成剪枝操作,同时利用粒子群算法实现对模型参数的优化。实验测试结果表明,所提算法的训练时间仅需7.8 s,性能与效率指标则均优于对比算法,且与人工评价法得到的指标大致相同,可以作为电力人才评价的辅助数据参考。

基于CN-FRAM的公共交通设备设施系统运营安全韧性度量

设备设施故障是公共交通系统运营安全事故发生的主要原因,为更好地度量和增强系统的安全韧性,提出融合复杂网络(CN)与功能共振分析方法(FRAM)的CN-FRAM运营安全韧性度量模型,并将系统韧性定义为扰动下系统性能损失与性能基线之比。首先,根据设备设施系统构成和节点功能,建立CN;其次,将FRAM模型嵌入到CN中,以扩展节点和连接,构建CN-FRAM模型;然后,基于CN-FRAM韧性度量模型分析系统组件之间功能变化的聚合,并在量化系统韧性时综合考虑网络整体效益和组件之间的耦合程度;最后,以南京市地铁信号系统为例,验证方法的可行性和有效性。结果表明:该模型可以量化系统破坏-恢复全过程的韧性,计算故障对系统的影响程度,并以韧性值最大化为目标,展现不同修复策略下的韧性表现,从而为确定最佳恢复顺序提供依据。对比现有方法,该方法所确定的最优恢复策略能显著减少系统因故障造成的整体性能损失,从而提高系统的韧性。

基于改进蚁群算法的中压配电网络规划方法研究

研究基于改进蚁群算法的中压配电网络规划方法,设计合理的中压配电网络规划方案,降低中压配电网络的费用。构建以负荷要求、潮流限制为约束条件,投资和运行费用最低为目标函数的中压配电网络规划数学模型。通过变换状态转移准则改进蚁群算法,以数学模型为基础采用改进蚁群算法规划中压配电网络方案。实验表明该方法能使中压配电网络的负荷处于合理范围内,降低投资与运行费用。

基于神经网络的智能电网稳定性预测模型研究

目前,在智能电网稳定性预测方面,机器学习发挥着越来越重要的作用。鉴于传统预测模型存在多种假设和预测不精确的缺点,提出了一种基于神经网络的智能电网稳定性预测模型,该模型以前馈神经网络为基础,采用考虑反应时间的阻尼最小二乘法对数据进行训练,将消耗、生产的有功功率和弹性系数作为输入变量,对表征智能电网稳定性的特征根实部进行预测,模型在隐藏层的激活函数采用双极性S函数(tansig),输出层的激活函数采用线性传递函数(purelin)。模型采用均方误差(MSE)和决定系数(R-Square,R2)对预测模型精确性和有效性进行评估。预测结果表明,该预测模型在训练和测试阶段均具有足够准确的预测性能,在预测范围具有极低MSE值和极大R2值,对不同潮流下智能电网稳定性的预测表现出了极高的准确性。

基于多岛遗传算法与响应面法的横向磁通感应加热装置参数优化设计

对金属带材进行横向磁通感应加热(TFIH)时,通常会存在加热温度分布不均匀以及加热温度偏离目标值两个问题。该文研究了加热器结构参数与电源参数对45号钢带材回火热处理温度的影响,并对两种参数分别进行优化,使带材在加热器出口处的平均温度达到目标值600℃,同时获得均匀的温度分布。采用Morris法对加热器结构参数进行全局灵敏度分析,选取显著影响相对不均匀度的参数并建立径向基函数(RBF)神经网络预测模型。使用多岛遗传算法(MIGA)对筛选的结构参数进行优化,初步获得均匀的温度分布。最后以降低温度分布的相对不均匀度和达到理想平均温度为目标,在优化后的加热器结构基础上使用响应面法(RSM)优化电源参数,实现多目标优化设计。仿真验证结果表明,45号钢带材在加热器出口处的平均温度为600.06℃、相对不均匀度为2.36%,满足45号钢回火热处理的要求。

基于平滑因子引入和神经网络优化的锂电池SOC估计方法

为提高锂电池荷电状态(SOC)的估计精度,提出了一种基于平滑因子引入和神经网络优化的锂电池SOC估计方法。将黄金分割优选法和模糊C均值聚类算法应用于RBF神经网络,分别用来确定最佳隐含层神经元个数和径向基中心;采用遗传算法对高斯核函数宽度及连接权值进行优化,解决了RBF神经网络结构和初始参数难以确定的问题。将滑动时间窗口内的放电容量作为平滑因子引入神经网络模型,增强了RBF网络对锂离子电池非线性特性拟合的能力。基于实验获得的锂离子电池在联邦城市行车计划(FUDS)工况下的数据,对所提出的方法进行仿真和验证,结果表明,所提方法显著提升了锂电池SOC的估计精度。

New Robust Exponential Stability Analysis for Uncertain Neural Networks with Time-varying Delay

In this paper,the global robust exponential stability is considered for a class of neural networks with parametric uncer- tainties and time-varying delay.By using Lyapunov functional method,and by resorting to the new technique for estimating the upper bound of the derivative of the Lyapunov functional,some less conservative exponential stability criteria are derived in terms of linear matrix inequalities (LMIs).Numerical examples are presented to show the effectiveness of the proposed method.

基于多时间尺度的电力系统源网协同自适应调度方法

在对电力系统源网进行调度时,受电力系统自身出力和负荷不确定性的影响,电力系统的运行性能难以达到理想效果,为此,提出基于多时间尺度的电力系统源网协同自适应调度方法。充分考虑由于天气环境的不确定性以及用户用电的不确定性导致的电力系统出力波动和负荷波动,在设置电力系统源网调度目标函数阶段,引入了第二个时间尺度——实时调度。在对目标函数进行求解时,利用区间可能度法,将电和冷热负荷等式约束转化为不等式约束的形式,利用反馈校正机制使得求解结果与电力系统实时状态形成信息闭环结构,实现对参数的滚动调度。在测试结果中,设计调度方法在提高电力系统运行性能方面具有良好的实际应用效果。

A data-derived soft-sensor method for monitoring effluent total phosphorus

The effluent total phosphorus(ETP) is an important parameter to evaluate the performance of wastewater treatment process(WWTP). In this study, a novel method, using a data-derived soft-sensor method, is proposed to obtain the reliable values of ETP online. First, a partial least square(PLS) method is introduced to select the related secondary variables of ETP based on the experimental data. Second, a radial basis function neural network(RBFNN) is developed to identify the relationship between the related secondary variables and ETP. This RBFNN easily optimizes the model parameters to improve the generalization ability of the soft-sensor. Finally, a monitoring system, based on the above PLS and RBFNN, named PLS-RBFNN-based soft-sensor system, is developed and tested in a real WWTP. Experimental results show that the proposed monitoring system can obtain the values of ETP online and own better predicting performance than some existing methods.

A RBF Network Learning Scheme Using Immune Algorithm Based on Information Entropy

A hybrid learning method combining immune algorithm and least square method is proposed to design the radial basis function(RBF) networks. The immune algorithm based on information entropy is used to determine the structure and parameters of RBF nonlinear hidden layer, and weights of RBF linear output layer are computed with least square method. By introducing the diversity control and immune memory mechanism, the algorithm improves the efficiency and overcomes the immature problem in genetic algorithm. Computer simulations demonstrate that the RBF networks designed in this method have fast convergence speed with good performances.

A global optimization algorithm based on multi-loop neural network control

This paper proposes an optimization algorithm based on a multi-loop control system with a neural network controller,in which the objective function that is used is the control plant of each sub-control system.To obtain the global optimization solution from a control plant that has many local minimum points,a transformation function is presented.On the one hand,this approach changes a complex objective function into a simple function under the condition of an unchanged globally optimal solution,to find the global optimization solution more easily by using a multi-loop control system.On the other hand,a special neural network(in which the node function can be simply positioned locally)that is composed of multiple transformation functions is used as the controller,which reduces the possibility of falling into local minimum points.At the same time,a filled function is presented as a control law;it can jump out of a local minimum point and move to another local minimum point that has a smaller value of the objective function.Finally,18 simulation examples are provided to show the effectiveness of the proposed method.