感应电机的神经网络鲁棒自适应动态面控制

针对感应电机的高阶、强耦合的非线性,设计一种具有自适应性和鲁棒性的动态面控制律,以实现对感应电机的控制。首先通过变量代换,将五阶感应电机模型变为块三角形式;接着采用神经网络逼近,将模型转化为含有结构不确定性和参数不确定性的参数化模型。对于感应电机系统中的参数不确定性,在动态面控制方法的基础上设计参数自适应律,在线估计其未知参数;对于感应电机系统中的结构不确定性,设计鲁棒控制律,使得电机控制系统具有鲁棒性。最后通过仿真实现,以验证所提出的控制方法的有效性与实用性。     

Decentralized robust adaptive neural dynamic surface control for multi‐machine excitation systems with static var compensator

Focusing on solving the control problem of the multimachine excitation systems with static var compensator (SVC), this paper proposes a decentralized neural adaptive dynamic surface control (DNADSC) scheme, where the radial basis function neural networks are used to approximate the unknown nonlinear dynamics of the subsystems and compensate the unknown nonlinear interactions. The main advantages of the proposed DNADSC scheme are summarized as follows: (1) the strong nonlinearities and complexities are mitigated when the SVC equipment are introduced to the multimachine excitation systems and the explosion of complexity problem of the backstepping method is overcome by combining the dynamic surface control method with neural networks (NNs) approximators; 2) the tracking error of the power angle can be kept in the prespecified performance curve by introducing the error transformed function; (3) instead of estimating the weighted vector itself, the norm of the weighted vector of the NNs are estimated, leading to the reduction of the computational burden. It is proved that all the signals in the multimachine excitation system with SVC are semiglobally uniformly ultimately bounded.     

基于自适应迭代学习的多智能体系统编队控制

针对带未知时变参数的非线性多智能体系统的编队问题,提出一种分布式自适应迭代学习控制策略。首先,通过傅里叶级数对系统的不确定参数进行展开,采用一个收敛级数序列处理傅里叶级数展开产生的截断误差,结合多智能体运行过程中的编队误差推导自适应迭代学习控制律和参数更新律;其次,针对领导者动态对大部分智能体都是未知的情况,设计新的辅助控制来补偿未知动态和避免未知有界干扰;然后,基于李亚普诺夫能量函数证明了在所设计控制律作用下多智能体系统编队误差随着迭代次数的增加在有限时间内趋于0;最后,将该控制策略运用到多无人机编队系统中,并通过搭建半物理实验平台,验证了控制方法的有效性。实验结果表明该控制方法可以确保多智能体快速形成所需编队,并且每个智能体在有限时间内可以精确跟踪期望轨迹。所提方法充分考虑了多智能体系统的参数不确定性以及抗干扰的能力,为实际应用中复杂多智能体系统的精确控制提供了有效的方法。     

遗忘因子随机配置网络驱动的自适应切换学习模型

随机配置网络(SCNs)具有通用逼近能力和快速建模特性,已成功应用于大数据分析。 在 SCN 的基础上,块增量随机配 置网络( BSC)使用块增量机制提高训练速度,但增加了模型结构的复杂程度。 为了解决上述难题,提出遗忘因子随机配置网 络(FSCN-I 和 FSCN-II)驱动的自适应切换学习模型(ASLM)。 该模型利用正态分布配置隐含层节点的输入参数。 FSCN-I 通过 误差值和遗忘因子调整节点块的尺寸,提高训练速度。 FSCN-II 引入节点移除机制降低模型结构的复杂程度。 ASLM 由 FSCN-I 和 FSCN-II 构成,两者根据自适应变化的边界随机切换以提高模型的训练速度,并在 FSCN-I 的基础上降低模型结构的复杂程 度。 最后,通过基础数据集和工业实例,表明该方法的有效性。     

基于深度学习的视频多目标码率分配策略

在高效视频编码（HEVC）中,最新的λ域码率控制方案自适应地对每个像素分配比特数,在自适应分配比特数的过程中,不考虑目标对象的编码质量。针对该问题提出了基于深度学习的视频多目标码率分配方案。在该方案中,首先采用目标检测算法SSD来检测识别编码视频中的多目标对象,获取对象信息。然后在帧级码率分配过程中,根据帧间差异,采用自适应比特比率,为每帧视频分配合理的比特数。最后,以更小的量化参数（QP）来提升目标对象编码质量。实验结果表明,与HM16.9相比,改进的算法在码控精度几乎不变的情况下,对全局的平均质量提升0.15 dB, 目标区域的质量提升0.35 dB。     

一种基于DBN的高光谱遥感图像分类方法

高光谱遥感数据越来越普及并为人们广泛使用,基于高光谱数据的地面物体精确分类是高光谱遥感技术的核心应用之一。对高光谱数据进行提特征提取是进行地物分类的有效方法。深度学习是机器学习研究中的新领域,它多隐层的多层感知器结构使其能够学习到对数据有更本质的刻画的特征,在图像分类和可视化领域取得了更好的成绩。深度置信网（deep belief network ,DBN）是深度学习网络中常见的模型。利用高光谱数据的高维特性,搭建基于DBN的高光谱图像分类模型,结合高光谱数据的空间结构对地物进行分类。实验表明,基于DBN的高光谱图像分类方法可以得到更好的分类效果。     

Numerical calculation of AC substation grounding systems buried in a vertical multilayered earth model by higher‐order basis function

To study the accuracy of numerical simulations for an AC substation grounding problem embedded in a vertical multilayered earth model, this paper proposes a novel algorithm combining the rapidly convergent one‐dimensional Galerkin's BEM with higher‐order basis functions on the basis of the quasi‐static electric field theory. General analytical formulas for the mutual impedance between any pair of short line conductors are also derived to accelerate the numerical integration. These formulas can work in not only complex space but also real space because of our application of the quasi‐static complex image method, which is different from the contribution of other earlier scientists, whose formula can work just in real space. The algorithm can be used to simulate and analyze any grounding system, including those incorporating a floating grid located anywhere in the earth model. Copyright © 2011 John Wiley & Sons, Ltd.     

自适应滤波算法的神经网络实现

为了提高传统自适应滤波器求解权值的速度,在Hopfield神经网络的基础上,提出自适应滤波算法的神经网络硬件实现,从理论上进行了分析,并进行了仿真。     

基于神经网络的一类非线性系统自适应滑模控制

针对可以分解为标称系统和不确定系统两部分的SISO非线性系统,提出了一种基于神经网络的自适应滑模控制方案。控制器由标称控制律和补偿控制律组成,标称控制律用来控制标称系统,补偿控制律是基于Lyapunov稳定性理论设计的自适应神经滑模控制律,用来控制不确定系统,而神经网络用来逼近系统的不确定性。理论分析和计算机仿真都证明,本文提出的控制策略不但能解决这类系统的轨迹跟踪控制问题,而且可以保证闭环系统的渐近稳定性。     

Adaptive steering of a containership based on neural networks

This paper describes the design of an adaptive autopilot for a containership. Well‐known results of linear adaptive theory are extended through the use of neural network to the case where the plant dynamics are described by a set of non‐linear equations. The desired behaviour of the containership during course‐changing manoeuvre is defined by the response of a reference model. Using an indirect approach, the controller parameters are adjusted based on the definition of an estimation error. The proposed adaptive controller is tested by a simulation study based on a non‐linear model, describing the dynamics of a containership in four degrees of freedom. In order to take into account the wave effects, a stochastic model of the ship wave‐induced motion is considered. Shallow water effect is also considered by changing some of the hydrodynamics parameters. Copyright © 2000 John Wiley & Sons, Ltd.     

Self‐recurrent wavelet neural network–based identification and adaptive predictive control of nonlinear dynamical systems

In this paper, the problem of simultaneous identification and predictive control of nonlinear dynamical systems using self‐recurrent wavelet neural network (SRWNN) is addressed. The structure of the SRWNN is a modification of the wavelet neural network (WNN). Unlike WNN, the neurons present in the hidden layer of SRWNN contain the weighted self‐feedback loops. Dynamic back‐propagation algorithm is employed to derive the necessary parameter update equations. To further improve the convergence speed of the parameters, a time‐varying (adaptive) learning rate is used. Four simulation examples are considered for testing the effectiveness of the proposed method. Furthermore, some disturbance rejection tests are also performed on the proposed method. The results obtained through the simulation study confirm the effectiveness of the proposed method.     

基于神经网络的混合速度辨识方法研究

提出了将开环直接计算与模型参考自适应方法相结合的神经网络混合转速辨识模型。仿真结果表明，基于该速度估计器的矢量控制系统动、静态性能好，解决了瞬时无功模型参考自适应方法在转速给定为负阶跃时的转速不稳定问题，转速估计精度高。     

多通道长短期记忆卷积网络的风速预测

为提高风速的预测性能,提出了多通道长短期记忆网络和卷积网络相结合的风速预测方法。预测模型由多个长短期记忆子网络及卷积网络组成。各子网络选择不同长度的历史数据作为输入,分别实现未来风速值的计算,避免了单一网络输入数据长度参数难以确定的问题。卷积网络将各子网络的计算结果进行卷积、最大池化操作,并通过全连接层计算风速序列的预测值。为避免预测误差累积及漂移,利用误差动态补偿方法对预测值进行校正,获得最终的预测结果。多通道长短期记忆卷积网络可用于风速的超短期预测中,仿真实验结果表明,与现有基于深度学习的预测网络相比,该网络能够更好地拟合实际风速序列的变化趋势,表现出更优的预测性能。     

基于神经网络辨识的模糊预测函数控制

针对生产过程中存在的滞后性、时变性、不确定性和变工况等特点及预测函数控制中模型失配的影响的情况,提出了基于神经网络辨识参数、通过模糊推理对控制量进行补偿的解决方案。并将基于神经网络辨识的模糊补偿预测函数控制应用于锅炉燃烧控制系统,通过连续系统仿真,结果表明这种控制器具有较强的鲁棒性。     

基于改进AdaBoost的油井动液面自适应集成建模

针对单一模型软测量算法在动液面预测应用过程中存在泛化能力弱、自适应性差等问题,提出利用Ada Boost集成学习的思想,突出预测误差在建模样本权重及弱学习机权重中的作用,使之更加适合回归模型预测。针对油井工况动态多变导致软测量模型随生产进行逐渐失效的问题,提出利用油田生产过程中定期巡检的少量动液面数据评估原有集成模型,利用相似度原理在保留原有模型信息的基础上增加新信息,并在此基础上根据新样本权重更新弱学习机模型,集成为强学习机模型以动态适应新的油田生产工况。通过对油田生产现场实际数据验证结果表明,该方法对油田生产波动的自适应能力强,能够提高动液面软测量模型的泛化能力及预测精度。     

Finite time adaptive fault-tolerant controller based on neural networks for switched stochastic nonlinear systems with actuator failures and disturbance

This article investigates the novel finite time adaptive neural fault-tolerant controller (FTC) for strict-feedback switched stochastic systems under arbitrary switching signals and takes into actuator failures including loss of effectiveness faults and bias faults consideration concurrently. Neural networks are utilized to approximate the unknown external disturbance and internal dynamics. On the basis of Itô differential equation and backstepping technique, an adaptive neural finite time FTC method is put forward. It is attested that the closed-loop systems are semiglobal practical finite time stable in probability and the tracking effects are great. Finally, to further demonstrate the high efficiency of proposed control method, two simulation examples are given.     

Identification and adaptive PID Control of a hexacopter UAV based on neural networks

In this paper, a novel adaptive PID controller for trajectory‐tracking tasks is proposed. It is implemented in discrete time over a hexacopter, and it takes into consideration the unmanned aerial vehicles (UAVs) nonlinear model. The PID controller is developed following an adaptive neural technique, and its stability is verified by the Lyapunov discrete theory. Besides, the neural identification of the dynamic model of the UAV is presented to backpropagate output errors to adjust PID gains with the purpose of reducing the control errors. The validation of the proposed algorithm is performed through experimental results with a hexacopter.     

柔性双工网络中基于图神经网络的能源效率联合优化方法

在绿色通信和双碳目标的背景下,提升网络能源效率(energy efficiency,EE)是当前无线通信系统设计和可靠运行的 关键技术之一。首先,针对柔性双工网络(flexible duplex networks,Flex-Net)的“和能源效率”最大化的问题,构建了一个具 有成对固定通信链路的网络;接着,鉴于图神经网络在通信网络资源优化中的优势,提出了一种新的基于图神经网络(GNN) 架构的柔性双工网络(GFlex-Net), 以联合优化“通信方向”和“传输功率”,达到网络能效最大化。仿真结果表明,与传统的算 法相比,所提出的架构有近乎最优的性能表现,达到穷举法性能的95%,但保持了较低的计算复杂度O(n²), 同时算法揭示了 GNN 在资源优化中面对样本复杂性、可扩展性和泛化能力方面的优势。     

基于小波神经网络的多节点机械手滑模自适应控制

由于多节点机械手的动力学模型比较复杂,难以直接精确得到,因此提出了以优化小波神经网络的 在线辨识算法为基础的一种稳定的自适应跟踪控制算法。并将滑模控制与自适应控制结合起来,确保了它 的稳定性和跟踪精度。仿真结果表明该算法的性能是令人满意的．     

基于小波神经网络的模拟电路故障诊断改进

在运用神经网络进行模拟电路故障诊断的过程中,代表着故障特征的网络输入至关重要,由于小波变换的时频局部化和多尺度分析等特性,将两者结合起来,通过小波变换对模拟电路的输出响应进行故障特征提取,同时解决PSPICE与MATLAB之间的数据通信问题,提出将蒙特卡罗分析产生的所有训练样本经过处理后输入到一个神经网络进行训练的方法,从而避免了训练多个神经网络。利用神经网络对各种故障模式进行分类,实现模拟电路的故障诊断,并进一步与传统的BP网络故障诊断法进行比较。仿真结果表明,该方法可以实现故障检测及定位,诊断的准确率显著提高,适用于模拟电路故障诊断。