基于神经网络同步学习的功率模块散热器拓扑优化快速迭代方法

随着功率模块集成化程度的提高,其散热结构优化已成为研发中的关键。拓扑优化可通过变换散热器形貌、结构来最大化地提升散热效果,因此受到了广泛关注。但在拓扑优化过程中,每步迭代均需要计算模块与散热器温度分布,占用较庞大的计算资源和计算时间。为加速传统散热器拓扑优化进程,在基于传统固体各向同性材料惩罚SIMP(solid isotropic material with penalization)散热器拓扑优化方法的基础上,提出一种嵌套神经网络NN(neural network)同步学习的快速迭代方法。首先,构建散热器基于编码器-解码器结构的NN预测模型,即基于散热器形貌迭代进化过程实现优化结构的快速预测;其次,将NN模型与散热器SIMP拓扑优化流程相嵌套,利用迭代过程中的中间形貌同步训练NN;最后,针对单芯片、两芯片模块结构,对比所提方法与传统迭代方法的拓扑优化结果,验证了所提NN同步学习方法的准确性和快速性。

基于自抗扰的H型平台模糊神经网络同步控制

为减小参数摄动、负载扰动、摩擦力等不确定扰动对H型平台中双直线电机位置同步精度的影响,提出一种自抗扰控制器(Active disturbance rejection controller,ADRC)和模糊型pi-sigma神经网络同步补偿器相结合的控制方法。采用自抗扰技术来观测未建模扰动和外界扰动,并将这些扰动视为系统的“总扰动”,实时给予补偿,以减小永磁直线同步电机(Permanent magnet linear synchronous motor,PMLSM)伺服系统的位置跟踪误差;同时,针对直驱H型平台双轴间的机械耦合和两台电机间参数动态不匹配的影响,采用模糊型pi-sigma神经网络同步补偿器以减小直驱H型平台的同步误差。最后,通过仿真验证,该控制策略可以有效提高H型平台的跟踪精度和同步精度,增强系统的抗扰性。

一个新的基于神经网络同步的公开密钥交换协议

首先介绍了在神经密码应用中的树型奇偶机模型,在综述神经密码协议研究的基础上,构建了一个使用含有衰减项的神经网络同步学习规则的新的密钥交换协议,并对其进行了理论分析。仿真实验结果表明,此协议可以使神经网络的同步性能提高4倍以上,对几种常见攻击的抵御表现良好,与传统的协议相比,新协议同步计算开销更小,安全性更高。

一种改进的基于树形奇偶机的神经网络同步方案

神经网络通过相互学习达到完全同步来构建密码协议,已经成为当今密码学的重要研究方向。针对神经密码同步过程中通信次数过多的问题,本文利用神经密码应用中的树形奇偶机,在综述神经密码协议研究的基础上,提出对神经网络单元初始权值的选取进行改进的解决方案。仿真实验结果表明,在保证安全性的情况下,本文方案大大加快了同步速度。

基于图卷积神经网络与K-means聚类的居民用户集群短期负荷预测

随着智能电表等高级量测装置在用户侧的广泛部署与使用,海量多源异构的居民用户数据得以采集与存储,为用户级负荷预测提供良好的数据基础。精准的居民用户集群负荷预测是促进智能配电网需求侧管理、辅助电网公司实现削峰填谷的重要基础。然而,现有的用户级负荷预测方法大多利用历史负荷序列的时间相关性构建数据驱动模型,却忽视相邻用户用电行为之间存在的潜在空间相关性。因此,提出一种基于K-means聚类和自适应时空同步图卷积神经网络的居民用户集群负荷预测方法。首先,采用K-means聚类将居民用户集群按照用电行为相似性划分成不同组;然后,基于居民用户集群的分组数量、各组居民用户的历史负荷数据以及各组居民用户负荷序列之间的相关性,构建面向居民用户集群负荷预测的时空图数据;最后,使用自适应时空同步图卷积神经网络实现居民用户集群短期负荷预测。文章通过真实的爱尔兰居民用户负荷公开数据集测试并验证所提方法的准确性和有效性,实验结果表明,相较于各个基准预测方法,所提方法能够充分挖掘并利用不同居民用户用电负荷之间的时空相关性,进而提高居民用户集群负荷预测精度。

基于初级视觉皮层的同步振荡目标检测模型

提出一种基于初级视觉皮层的目标检测模型。该模型只采用方位选择性细胞和皮层内水平连接等V1基本单元 ,它以链码表示的目标轮廓作为知识 ,允许该知识以时间脉冲的形式控制V1区内神经细胞的动态活动 ,使与知识轮廓形状相符合的轮廓内的细胞进入同步振荡状态 ,实现对视野中特定目标轮廓的识别。计算机仿真结果表明 ,在较高级皮层的“知识”控制之下 ,初级视觉皮层结构上实现简单的目标检测是可行的。

基于SET-CNN的紧凑型地波雷达弱目标检测方法

紧凑型高频地波雷达发射功率较低且采用小孔径阵列,导致雷达回波中的弱目标增多,进而引起基于距离-多普勒谱的目标检测方法性能降低,目标探测能力减弱。为提高紧凑型地波雷达对弱目标的检测性能,本文提出一种基于同步提取变换-卷积神经网络(Synchroextracting Transform-Convolutional Neural Network,SET-CNN)的紧凑型地波雷达弱目标检测方法:首先在时频谱处理中,利用信噪比方法抑制信号中的海杂波,减少杂波时频脊线对目标检测的影响;然后基于SET时频谱构建时频脊线样本数据库,再通过卷积神经网络进行时频脊线分类,并基于分类结果的后处理完成船只目标检测。通过仿真和实测数据验证提出的目标检测方法,结果表明,本文提出的方法能够有效检测到弱目标,提高紧凑型地波雷达的目标检测性能。

光照不变量特征提取新方法及其在目标识别中的应用

针对LNSCT光照不变量提取方法因舍弃低频分量而丢失目标轮廓信息的问题,本文提出了一种新的光照不变量提取方法 MLNCST.新方法首先用NSCT将对数域的输入图像进行第一重多尺度分解,实现低频分量和高频分量的分离;其次对高频子带系数进行Bayes Shrink阈值滤波,低频分量做逆NSCT得到其特征图像;然后对特征图像进行第二重NSCT分解,并对分解后的高频子带阈值滤波以及低频分量逆NSCT;经多重NSCT分解,最后由多次分解后的高频子带系数集提取光照不变量特征.经进一步研究光照不变量特征与原始图像之间的关系,设计了并行同步卷积神经网络-Dual Lenet,通过融合两者的高层特征来提高地面目标识别的准确率.实验结果显示,在Lenet模型下,MLNSCT比LNSCT具有更高的分类准确率,并且随着分解重数的增加分类准确率更高;同时融合了光照不变量特征的Dual lenet能进一步提高地面目标识别准确率.

Improved results on synchronization in arrays of coupled delayed neural networks with hybrid coupling

In order to investigate the influence of hybrid coupling on the synchronization of delayed neural networks, by choosing an improved delay-dependent Lyapunov-Krasovskii functional, one less conservative asymptotical criterion based on linear matrix inequality （LMI） is established. The Kronecker product and convex combination techniques are employed. Also the bounds of time-varying delays and delay derivatives are fully considered. By adjusting the inner coupling matrix parameters and using the Matlab LMI toolbox, the design and applications of addressed coupled networks can be realized. Finally, the efficiency and applicability of the proposed results are illustrated by a numerical example with simulations.

An adaptive blind watermarking scheme utilizing neural network for synchronization

An important problem constraining the practical implementation of robust watermarking technology is the low robustness of existing algorithms against geometrical distortions. An adaptive blind watermarking scheme utilizing neural network for synchronization is proposed in this paper,which allows to recover watermark even if the image has been subjected to generalized geometrical transforms. Through classification of image’s brightness, texture and contrast sensitivity utilizing fuzzy clustering theory and human visual system, more robust watermark is adaptively embedded in DWT domain. In order to register rotation, scaling and translation parameters, feedforward neural network is utilized to learn image geometric pattern represented by six combined low order image moments. The distortion can be inverted after determining the affine distortion applied to the image and watermark can be extracted in a standard way without original image. It only needs a trained neural network. Experimental results demonstrate its advantages over previous method in terms of computational effectiveness and parameter estimation accuracy. It can embed more robust watermark under certain visual distance, and effectively resist JPEG compression, noise and geometric attacks.

Exponential synchronization of general chaotic delayed neural networks via hybrid feedback

This paper investigates the exponential synchronization problem of some chaotic delayed neural networks based on the proposed general neural network model,which is the interconnection of a linear delayed dynamic system and a bounded static nonlinear operator,and covers several well-known neural networks,such as Hopfield neural networks,cellular neural networks(CNNs),bidirectional associative memory(BAM)networks,recurrent multilayer perceptrons(RMLPs).By virtue of Lyapunov-Krasovskii stability theory and linear matrix inequality(LMI)technique,some exponential synchronization criteria are derived.Using the drive-response concept,hybrid feedback controllers are designed to synchronize two identical chaotic neural networks based on those synchronization criteria.Finally,detailed comparisons with existing results are made and numerical simulations are carried out to demonstrate the effectiveness of the established synchronization laws.

Coherence Resonance and Noise-Induced Synchronization in Hindmarsh-Rose Neural Network with Different Topologies

In this paper, we investigate coherence resonance （CR） and noise-induced synchronization in Hindmarsh- Rose （HR） neural network with three different types of topologies： regular, random, and small-world. It is found that the additive noise can induce CR in HR neural network with different topologies and its coherence is optimized by a proper noise level. It is also found that as coupling strength increases the plateau in the measure of coherence curve becomes broadened and the effects of network topology is more pronounced simultaneously. Moreover, we find that increasing the probability p of the network topology leads to an enhancement of noise-induced synchronization in HR neurons network.

Synchronization of a Class of Delayed Neural Networks with Sector Nonlinearity

In this paper, global synchronization is discussed for a general class of delayed neural networks with time-varying and distributed delays. Furthermore, the activation func- tions in the neural networks can be different type. Based on the drive-response concept and the Lyapunov stability theorem, some sufficient criteria are obtained to guarantee the global synchronization of the considered models even when input sector nonlinearity caused by physical limitations is presented in response systems. Finally, a typical example is also given to illustrate the effectiveness of the proposed synchronization scheme.

Approach to Generalized Synchronization with Application to Chaos-Based Secure Communication

A constructive theorem is established for generalized synchronization (GS) related to C1 diffeomorphic transformations of unidirectionally coupled dynamical arrays. The theorem provides some interpretations about the underlying mechanism of various GS phenomena in nature. As a direct application of the theorem, a chaos-based secure Internet communication scheme is proposed. Moreover, a cellular neural network (CNN) of Chen's chaotic circuits with GS property is designed and studied. Numerical simulation shows that this Chen's CNN has high security and is fast and reliable for secure Internet communications.

Phenomenological Simulation Study of Neuronal Activity Synchronization in 110 Elements Network

The phenomenon of activity synchronization in biological neural network is considered. Simulation of neurons dynamics in the 6-layer neural network with 110 elements in different regimes： regular spikes, chaotic spikes, regular and chaotic bursting, etc was performed. Izhykevich＇s phenomenological model that displays different types of activity inherent for real biological neurons was used for simulation. Space-time diagram for the entire network and raster plots for the whole structure and for each layer separately were built for visual inspection of neural network activity synchronization. Synchronization coefficients based on cross-correlation times of action potentials for all neurons pairs were calculated for the whole neural system and for each layer separately.

Image Segmentation Based on Period Difference of the Oscillation

A new method for image segmentation based on pulse neural network is proposed. Every neuron in the network represents one pixel in the image and the network is locally connected. Each group of the neurons that correspond to each object synchronizes while different groups of the neurons oscillate at different period. Applying this period difference, different objects are divided. In addition to simulation, an analysis of the mechanism of the method is presented in this paper.

Image Edge Detection Based on Oscillation

A new method for image edge detection based on a pulse neural network is proposed in this paper. The network is locally connected. The external input of each neuron of the network is gray value of the corresponding pixel. The synchrony of the neuron and its neighbors is detected by detection neurons. The edge of the image can be read off at minima of the total activity of the detection neurons.

Artificial Neural Network in Harmonic Reduction of STATCOM

To eliminate harmonic pollution incurred from the static synchronous compensator(STATCOM), a method of applying artificial neural network is presented. When PWM wave is formed based on the harmonic suppression theory, a concave is set on certain angle of the square wave to suppress unnecessary harmonics, by timely and on-line determining the chopping angle corresponding to respective harmonics through artificial neural network, i.e. by setting the position of concave to eliminate corresponding harmonics, the harmonic component on output voltage of the inverter can be improved. To conclude through computer simulation test, the perfect control effect has been proved.

Effects of Coupling Distance on Synchronization and Coherence in Chaotic Neural Networks

Effects of coupling distance on synchronization and coherence of chaotic neurons in complex networks arenumerically investigated.We find that it is not beneficial to neurons synchronization if confining the coupling distanceof random edges to a limit d_(max),but help to improve their coherence.Moreover,there is an optimal value of d_(max) atwhich the coherence is maximum.

A novel license plate recognition method using HTD and VTD features

In this paper, a novel method of licence plate recognition (LPR) using the vertical traverse density (VTD) and horizontal traverse density (HTD) is presented. The neutral network algorithm using VTD and HTD features is also an innovation. In addition, a so called secondary recognition method which splits characters into different parts is developed. Experimental results show that it is a simple and fast algorithm, which meets the request of real time and nicety performances of LPR and thus has applied value in intelligence transportation system (ITS).