A generalized synthesis load model considering network parameters and allvanadium redox flow battery

The simulation precision of the classic load model(CLM)is affected by the increasing proportion of installed energy storage capacity in the grid.This paper studies the all-vanadium redox flow battery(VRB)and proposes an equivalent model based on the measurement-based load modeling method,which can simulate the maximum output of the VRB energy storage system and fit the external characteristic of the system precisely in the occurrence of large disturbance and continuous small disturbance.The equivalent model is connected to CLM to form a generalized synthesis load model(GSLM),which considers the parameters of distribution network and reactive power compensation.Compared with CLM,GSLM has better structures and can describe the load characteristics of distribution network with energy storage system more precisely.Simulation results validate the effectiveness and good parameter stability of GSLM,and show that the higher the proportion of energy storage in the grid is the better description ability GSLM has.

A back-propagation neural-network-based displacement back analysis for the identification of the geomechanical parameters of the Yonglang landslide in China

Xigeda formation is a type of hundredmeter-thick lacustrine sediments of being prone to triggering landslides along the trunk channel and tributaries of the upper Yangtze River in China. The Yonglang landslide located near Yonglang Town of Dechang County in Sichuan Province of China, which was a typical Xigeda formation landslide, was stabilized by anti-slide piles. Loading tests on a loading-test pile were conducted to measure the displacements and moments. The uncertainty of the tested geomechanical parameters of the Yonglang landslide over certain ranges would be problematic during the evaluation of the landslide. Thus, uniform design was introduced in the experimental design,and by which, numerical analyses of the loading-test pile were performed using Fast Lagrangian Analysis of Continua(FLAC3D) to acquire a database of the geomechanical parameters of the Yonglang landslide and the corresponding displacements of the loadingtest pile. A three-layer back-propagation neural network was established and trained with the database, and then tested and verified for its accuracy and reliability in numerical simulations. Displacement back analysis was conducted by substituting the displacements of the loading-test pile to the well-trained three-layer back-propagation neural network so as to identify the geomechanical parameters of the Yonglang landslide. The neuralnetwork-based displacement back analysis method with the proposed methodology is verified to be accurate and reliable for the identification of the uncertain geomechanical parameters of landslides.

Prediction of Superconductivity for Oxides Based on Structural Parameters and Artificial Neural Network Method

Superconductive properties for oxides were predicted by artificial neural network (ANN) method with structural and chemical parameters as inputs. The predicted properties include superconductivity for oxides, distributed ranges of the superconductive transition temperature (Tc) for complex oxides, and Tc values for cuprate superconductors. The calculated results indicated that the adjusted ANN can be used to predict superconductive properties for unknown oxides.

PARAMETERS INVERSION OF FLUID-SATURATED POROUS MEDIA BASED ON NEURAL NETWORKS

The multi- layers feedforward neural network is used for inversion ofmaterial constants of fluid-saturated porous media. The direct analysis of fluid-saturated porousmedia is carried out with the boundary element method. The dynamic displacement responses obtainedfrom direct analysis for prescribed material parameters constitute the sample sets training neuralnetwork. By virtue of the effective L-M training algorithm and the Tikhonov regularization method aswell as the GCV method for an appropriate selection of regu-larization parameter, the inversemapping from dynamic displacement responses to material constants is performed. Numerical examplesdemonstrate the validity of the neural network method.

轻量级重参数化的遥感图像超分辨率重建网络设计

针对当前基于深度学习的遥感图像超分辨率重建模型部署时对硬件要求较高,本文设计了一种轻量级基于重参数化的残差特征遥感图像超分辨率重建网络。首先,采用重参数化方法设计了一种残差局部特征模块,以有效地提取图像局部特征;同时考虑到图像内部出现的相似特征,设计了一个轻量级的全局上下文模块对图像的相似特征进行关联以提升网络的特征表达能力,并通过调整该模块的通道压缩倍数来减少模型的参数量和改善模型的性能;最后,在上采样模块前使用多层特征融合模块聚合所有的深度特征,以产生更全面的特征表示。在UC Merced遥感数据集上进行测试,该算法在遥感图像3倍超分辨率下的参数量为539 K,峰值信噪比为30.01 dB,结构相似性为0.8449,模型的推理时间为0.010 s;而HSENet算法的参数量为5470 K,峰值信噪比为30.00 dB,结构相似性为0.8420,模型的推理时间为0.059 s。实验结果表明,该算法相比HSENet算法,参数量更少,运行速度较快,且峰值信噪比与结构相似性也有一定的提高。在DIV2K自然图像数据集上进行测试,该算法的峰值信噪比和结构相似性相比其他算法也有一定的优势,表明该算法的泛化能力较强。

Aircraft parameter estimation using a stacked long short-term memory network and Levenberg-Marquardt method

To effectively estimate the unknown aerodynamic parameters from the aircraft’s flight data,this paper proposes a novel aerodynamic parameter estimation method incorporating a stacked Long Short-Term Memory(LSTM)network model and the Levenberg-Marquardt(LM)method.The stacked LSTM network model was designed to realize the aircraft dynamics modeling by utilizing a frame of nonlinear functional mapping based entirely on the measured input-output data of the aircraft system without requiring explicit postulation of the dynamics.The LM method combines the already-trained LSTM network model to optimize the unknown aerodynamic parameters.The proposed method is applied by using the real flight data,generated by ATTAS aircraft and a bio-inspired morphing Unmanned Aerial Vehicle(UAV).The investigation reveals that for the two different flight data,the designed stacked LSTM network structure can maintain the efficacy of the network prediction capability only by appropriately adjusting the dropout rates of its hidden layers without changing other network parameters(i.e.,the initial weights,initial biases,number of hidden cells,time-steps,learning rate,and number of training iterations).Besides,the proposed method’s effectiveness and potential are demonstrated by comparing the estimated results of the ATTAS aircraft or the bio-inspired morphing UAV with the corresponding reference values or wind-tunnel results.

基于改进RT-DETR的路面坑槽检测模型

路面坑槽对驾驶的舒适性和安全性有很大影响。针对路面图像中坑槽尺寸小和特征信息匮乏导致检测精度低的问题,提出一种基于RT-DETR的路面坑槽检测模型Pavement Pothole-DETR(PP-DETR)。其主干网络使用SPDRSFE模块进行特征提取,可保留更多特征信息,提高小目标检测精度;引入渐进特征金字塔网络实现特征融合,避免多级传输造成的信息丢失,以解决坑槽特征信息主要集中在中、底特征层的问题;使用结构重参数化模块Conv3XCC3进行特征再提取,在提高网络表达能力的同时又不增加计算量。实验结果显示,相比原RT-DETR模型,PP-DETR的精确率与召回率分别提升了2.9和5.4个百分点,mAP达到76.9%。本文提出的改进方法有效提升了网络的特征提取和特征融合能力,在路面坑槽检测任务上的表现明显优于YOLO系列模型。

基于双线性RepVGG注意力网络的花卉分类

为进一步提高花卉分类的准确率,在对双线性卷积神经网络、RepVGG及注意力机制进行研究的基础上,提出一种基于双线性RepVGG注意力机制的网络模型。首先利用RepVGG网络替换原始的特征提取网络VGG,以提高对花卉主要特征的提取能力;然后在两个RepVGG网络中分别引入通道注意力及空间注意力机制,并利用两个RepVGG网络外积后生成的高维双线性特征,来提取花卉的细粒度特征;最后通过结构重参数化,将RepVGG的各层转换为单路结构,以提高模型推理的速度。实验结果表明,在增强的Oxford-102数据集上,新模型与原始模型及常见模型相比,其推理速度及分类准确率均有较大的提升,与未引入注意力前相比,分类准确率也有一定的提升。

基于改进YOLOv5s的滚动轴承表面缺陷识别算法

为了解决机械设备轴承表面缺陷检测中多目标情形下的小目标漏检率高、检测速度慢、模型精度和特征提取泛化能力不足的问题,提出一种滚动轴承表面缺陷识别网络模型YOLOv5s-CDOD。在卷积操作前,使用B-ConvNeXt网络平衡模型的精度和复杂度,保留轴承表面小目标缺陷的特征,同时提升模型的泛化能力;通过将YOLOv5s网络中的传统卷积模块替换为具有二次深度过参数化卷积的卷积(DOD-Conv)模块,在不增加模型参数的情况下,提高模型的识别精度和速度;最后,在特征处理阶段,使用VariFocal Loss损失函数,增加模型对正样本目标的学习,对轴承小目标缺陷的检测精度进一步提升。实验结果表明:与原YOLOv5s网络相比,优化后的网络参数量减少了10%,使得模型的检测速度明显提升;同时,所提模型的平均检测精度达到了94%,对轴承表面小目标缺陷的识别率也有所提高。

基于重参数化的超分辨率重建

针对现有单图像超分辨率重建(Single Image Super-resolution,SISR)模型存在速度和精度的矛盾,论文给出了一种重参数化(Re-parameterization)的轻量模型用于实现图像重建。该模型训练时通过使用结构较复杂的模型保证精度,推理时通过模型等效变换为简单的卷积以提高速度。同时多监督结构的加入让模型收敛更快且更为灵活。通过峰值信噪比和结构相似度对重建模型的质量和效率进行了评估。验证了所提模型在现有超分辨率重建方法中兼具了轻量和重建质量良好的优点。

基于重参数化的注意力机制算法

为提高注意力机制对深度神经网络准确率的提升效果,提出一种重参数化通道注意力模块(RCAM)。鉴于挤压激励网络的通道压缩方法对网络精度存在较大误差,故提出一种基于重参数化技术的通道重参数化模块,将此模块与注意力机制进行有效结合;按集成策略消融实验所获得的结果,将此注意力模块放置进主干网络中。实验结果表明,在公共数据集CIFAR-100和ImageNet-100,主干网络为RepVGG_A0、ResNet-18时,其准确率分别较未添加注意力机制的网络提升了2.37%和1.72%,1.61%和0.71%,将结果与其它注意力机制进行比较,验证了基于重参数化的注意力机制对主干网络的提升远优于其它方法。

Improved Implementation and Evaluation of Wireless Sensor Networks in Intelligent Building

A complete study for the implementation of wireless sensor networks in the intelligent building is presented. We carry out some experiments to find out the factors affecting the network performance. Several vital parameters which are related to the link quality are measured before deploying the actual system. And then, we propose an optimized routing protocol based on the analysis of the test data. We evaluate the deployment strategies to ensure the excellent performance of the wireless sensor networks under the real working conditions. And the evaluation results show that the presented system could satisfy the requirements of the applications in the intelligent building.

过载和攻击时间约束下的非线性最优制导方法

考虑导弹过载受限条件下,对以期望时间为攻击目标的非线性最优制导问题进行了研究。首先,建立了非线性最优制导问题的理论模型,基于庞特里亚金极大值原理和饱和函数方法建立了最优轨迹的最优性条件。其次,根据最优性条件和哈密尔顿轨迹参数化方法,建立了最优轨迹的参数化微分方程组,使得通过数值积分即可生成从飞行状态到最优制导指令映射关系的数据集。然后,通过前馈神经网络对上述映射关系进行近似,实现了非线性最优制导指令的毫秒量级实时生成。最后,通过数值仿真验证了所提非线性最优制导指令生成方法的有效性。

基于ANN模型的内冷型溶液除湿器性能研究

溶液除湿器因可被低品位热能驱动,且具有除湿效率高等优点而受到广泛关注,但其传质性能的预测准确度还有待提高.本文搭建了单通道内冷型溶液除湿实验平台,研究了不同参数对于除湿过程中传质性能的影响,同时,建立了基于MATLAB平台的人工神经网络(ANN)模型用于预测传质性能,并用上述实验数据对该ANN模型进行了验证.结果表明,ANN模型预测得出的Sh与实验Sh平均绝对相对偏差(MARD)为4.07%.与现有经验公式相比,建立的ANN模型预测精度更高.此外,还利用ANN模型研究了不同参数变化下的Sh的变化趋势,从而分析不同参数对除湿性能的影响.

考虑自动驾驶仪延迟的非线性最优末制导方法

针对过载受限和自动驾驶仪延迟等条件下的非线性最优末制导指令在线生成方法进行研究。首先,基于庞特里亚金极大值原理,建立了目标静止的非线性最优末制导问题的最优性条件,并利用饱和函数将过载约束嵌入最优性条件。其次,应用参数化方法使得通过数值积分即可快速生成满足最优性条件的飞行轨迹数据集。然后,利用该数据集训练神经网络,使其拟合弹-目相对运动状态到最优制导指令的映射关系,实现过载约束下制导指令的毫秒量级在线生成。针对自动驾驶仪的延迟响应,通过微分补偿法估计神经网络下一时刻输出的制导指令以实现快速跟踪。最后,仿真结果表明,本文所提出方法针对静止目标与小机动目标都能够在线生成最优制导指令。

基于结构重参数化的权重嵌套式教师-学生网络

教师-学生网络在知识蒸馏、知识扩展、域自适应以及多任务学习等领域的应用中均已取得显著成果且成为重要的框架或范式。在不同的应用场景中,教师-学生网络拥有不同的优化方向。本文针对半监督应用场景下的知识扩展任务提出了一种权重嵌套式教师-学生网络模型。该模型借鉴结构重参数化方法,将学生网络设计为结构重参数化网络,并进一步以其主干分支作为教师网络,实现教师网络和学生网络的嵌套。由于教师网络和学生网络的主干分支共享权重,因此本文所提出的框架可以大幅地缩减传统教师-学生网络模型的参数量以及对显存的要求。为了验证本文提出的框架在半监督设定下的有效性,我们在Food-101数据集上进行了对比实验。实验结果表明嵌套式的教师-学生网络在模型参数量和显存占用量减少超过40%的条件下,实现了在多种标注比例下均取得略高于传统教师-学生网络模型的性能。

面向鲁棒图结构防御的过参数化图神经网络

图数据在现实应用中普遍存在,图神经网络(GNN)被广泛应用于分析图数据,然而GNN的性能会被图结构上的对抗攻击剧烈影响.应对图结构上的对抗攻击,现有的防御方法一般基于图内聚先验进行低秩图结构重构.但是现有的图结构对抗防御方法无法自适应秩真值进行低秩图结构重构,同时低秩图结构与下游任务语义存在错配.为了解决以上问题,基于过参数化的隐式正则效应提出过参数化图神经网络(OPGNN)方法,并形式化证明所提方法可以自适应求解低秩图结构,同时证明节点深层表征上的过参数化残差链接可以有效解决语义错配.在真实数据集上的实验结果表明,OPGNN方法相对于现有基线方法具有更好的鲁棒性,同时,OPGNN方法框架在不同的图神经网络骨干上如GCN、APPNP和GPRGNN上显著有效.

基于深度学习的翼型参数化建模方法

为解决现有翼型几何参数化描述方法优化设计效率低、计算工作量大的问题,提出了一种基于深度学习的翼型参数化建模方法。该方法以伊利诺伊大学厄巴纳-香槟分校(University of Illinois at Urbana-Champaign,UIUC)翼型数据库中翼型上下表面坐标点转化的翼型二维图像作为输入,首先使用卷积运算提取大量翼型图像的几何特征,然后通过多层感知机对提取的几何特征进行分类和压缩,将翼型形状压缩成若干个简化的拟合参数,最后通过解码器恢复翼型图像并输出翼型上下表面的点坐标。在此基础上,探讨了拟合参数数量对翼型几何精度的影响,确定了含6个拟合参数的卷积神经网络(convolutional neural network,CNN)结构,并基于计算流体力学数值仿真验证了所提出方法的拟合精度。最后,开发了可视化翼型几何设计软件,实现了拟合参数的调整与修正,并分析了各拟合参数对翼型形状的影响规律。结果表明,6个拟合参数均会对翼型形状产生全局影响,单独或联合调整6个拟合参数可获得新的翼型设计空间。研究结果可为翼型的优化设计提供技术支持与理论参考。

PARAMETER COORDINATION AND ROBUST OPTIMIZATION FOR MULTIDISCIPLINARY DESIGN

A new parameter coordination and robust optimization approach for multidisciplinary design is presented. Firstly, the constraints network model is established to support engineering change, coordination and optimization. In this model, interval boxes are adopted to describe the uncertainty of design parameters quantitatively to enhance the design robustness. Secondly, the parameter coordination method is presented to solve the constraints network model, monitor the potential conflicts due to engineering changes, and obtain the consistency solution space corresponding to the given product specifications. Finally, the robust parameter optimization model is established, and genetic arithmetic is used to obtain the robust optimization parameter. An example of bogie design is analyzed to show the scheme to be effective.

基于国产FPGA的移动通信网信令设计与实现

为提高电信网设备应对异常信令访问的检测能力,需对64K信令进行分析并处理。为了提高解析效率并满足近年来相关产品对自主可控越来越高的要求,设计了一种基于国产现场可编程门阵列(Field Programmable Gate Array, FPGA)的信令解析方案,给出了方案的总体设计思路,并对FPGA实现的功能模块进行详细说明。对系统进行设计时,采用模块化参数化方法以及在关键环节添加状态参数,提高了可扩展性并可以对模块内部运行状态进行监控,最终实现了对信令高效且灵活的解析,主要器件等均为国产。经过测试,可以实现STM-1(STM-Synchronous Transfer Module-1)数据的接入、串并转换、HDLC(High-level Data Link Control)解帧等功能,完成32路64K信令的并发处理,模块运行状态可查可看,达到了预期的效果。以STM-1为例,基于现有功能的模块化设计,可以平滑地扩展到STM-4、STM-16的应用。