Due to advancements in the manufacturing techniques of transformer core materials,the precision of traditional excitation inrush current detection methods has progressively diminished,thereby precipitating the issue of erroneous operation in transformer differential protection systems.To address this issue,a novel CNN-BiLSTM excitation inrush current identification model incorporating an attention mechanism was proposed.Initially,various instantaneous current data were collected using the PSCAD simulation platform,simulating transformer energization,parallel operation,and internal faults,forming the dataset for the model.Subsequently,the CNN-BiLSTM-Attention model was constructed and trained on MATLAB.Finally,comparative experiments are conducted using different models,followed by an analysis of noise resistance.Results demonstrate that the CNN-BiLSTM-Attention neural network not only distinguishes excitation inrush,over-excitation,and fault currents with 100%accuracy,but also exhibits superior differentiation among different fault currents.The model has been shown to exhibit strengths including robust generalization performance,high fitting capacity,and resilient noise resistance.
In modern power systems, the integration of renewable energy sources exacerbates voltage sag issues, which significantly impact power quality. This paper proposes a novel voltage sag measurement method that combines a root mean square (RMS) sliding window difference operator with sampled sequence reconstruction to address the limitations of traditional detection methods. Unlike conventional approaches, the proposed method accurately identifies the start and end times of voltage sag using a predefined threshold, thereby ensuring precise interval isolation for detailed analysis. Moreover, phase jump detection is enhanced through sampled sequence reconstruction. These enable the accurate extraction of voltage sag characteristics, including depth, duration, and phase jump. The simplicity and computational efficiency of the algorithm make it highly suitable for implementation in embedded systems. Comprehensive simulations and real-world measurements have been conducted to validate the effectiveness of the proposed method, demonstrating significant improvements in voltage sag detection accuracy and robustness against noise and other disturbances.
In the reliability analysis of electrical integrated energy systems,traditional numerical algorithms struggle with the high computational demands of natural gas dynamics,making timely dynamic analysis difficult.This paper presents a novel approach that replaces these algorithms with a neural network-based method using multi-scale dilated convolution and attention mechanisms.The model utilizes convolutional neural networks(CNN)for feature extraction and long short-term memory(LSTM)networks to capture time series characteristics.Multi-scale dilated convolutions expand the receptive field,while attention mechanisms enhance sensitivity to critical changes.This sequence-to-sequence learning process accurately models complex relationships between time steps,resulting in a dynamic surrogate model for the gas network.The gas network model is integrated with the power system flow model,allowing for a comprehensive reliability analysis using Monte Carlo methods and multi-state models.Tests on a distribution-level electric-gas integrated energy system show that the CNN-LSTM model not only accurately simulates gas dynamics but also significantly improves computational efficiency,meeting the reliability assessment needs of large-scale integrated energy systems.
This study addresses the limitations of existing indirect bridge damage identification methods that are based on the vehicle-bridge coupled vibration theory of highway bridges. To overcome these shortcomings, we propose an extended approach that incorporates various types of deep-learning models with vehicle-bridge coupled vibration responses. The proposed method is demonstrated using a three-span continuous beam bridge as a case study. First, a vehicle and bridge analysis model is established, and bridge damage is simulated using unit stiffness reduction, considering different damage scenarios. Next, to account for road roughness randomness, vehicle-bridge coupling vibration analysis is performed under various road roughness conditions, yielding the vertical acceleration vibration signal of the vehicle. Subsequently, we employ an end-to-end damage recognition method, utilizing the vehicle acceleration response as the network input, to construct two types of deep-learning models: one-dimensional convolutional neural network (1D-CNN) and convolutional long short-term memory neural network (CNN-LSTM). The recognition performance of both models is compared and analyzed. Taking Zhengzhou Taohuayu Self-Anchored Suspension Bridge in China as an example, this study delves into the capability of bridge damage identification using deep learning. The results demonstrate that the one-dimensional convolutional neural network achieves excellent recognition performance in terms of both damage location and severity.
Current infrared and visible image fusion algorithms often suffer from issues such as unclear texture details in the fused image and an unbalanced display of infrared information and texture details.In this paper,we propose an image fusion method of pseudo-color infrared and visible images based on attention-dense network.The greyscale infrared image is first processed in pseudo-color and then combined with the colored visible image to form a multi-channel data input fusion network.Secondly,a generator network structure consisting of convolutional layers and densely connected blocks with attention modules is designed to focus on the key information of the source image and enhance the ability of the network to extract information from the source image.Finally,the content loss function is constructed by using infrared pixels,visible pixels,visible gradient and infrared gradient to keep the stability of infrared target and texture details in the fused image.Qualitative and quantitative comparisons are made with five representative fusion methods.The results show that the peak signal-to-noise ratio,information entropy,average gradient,and mutual information of the fused images obtained by this method achieve the optimal values of 31.684 1,6.558 1,6.009 6,and 3.096 0,respectively.The quantitative and qualitative results demonstrate that the proposed fusion method results in a fused image with richer texture details and good visual effects.
近年来,随着人工智能技术的发展,可独立完成任务的自主智能机器人成为研究热点.视觉感知和步态控制作为机器人完成各项任务的重要手段,是亟需不断深入研究的技术点.为了拓展小型人形机器人的自主智能水平和应用场景,本文从国际自主智能机器人大赛技术问题出发,研究分析视觉感知和步态控制在模拟人类生活场景的任务(即U型赛)中的应用.首先,对比赛环境和软硬件平台进行介绍,其次分别分析了视觉感知和步态控制在大赛中的应用方法和算法实现,并以上下开横杆为例,介绍了 U型赛参赛队伍所使用的算法并进行分析比较.随后,展示了历年参赛队伍的优秀成果,分析了比赛任务完成度,指出比赛的研究难点及可能的解决方向.最后,对未来大赛的技术应用发展方向进行总结.
It is very important to determine and master the exercise intensity in exercise. Appropriate exercise intensity can effectively improve the body function. Too much intensity can make the body function decline, and even damage the health. Aiming at the problems of few assessment indicators and low accuracy of traditional exercise intensity assessment methods, this paper proposed an exercise intensity assessment method based on knowledge graph. Firstly, the indications of exercise intensity assessment were extracted by BERT-CRF, and the knowledge graph of exercise intensity assessment was established in the Neo4j database. Secondly, on the basis of RecGNNs knowledge reasoning, personalized and accurate assessment of exercise in-tensity for different age groups was realized. The exercise intensity assessment system based on this method can display the user’s exercise data index and personalized exercise intensity assessment results. Finally, the exercise intensity assessment system was constructed. The experimental results show that the exercise intensity assessment system based on knowledge graph was stable and reliable, and the accuracy of personalized assessment can reach 85.7%.
为解决稀疏立足地形中单腿失效六足机器人无法继续稳定行走的问题,提出一种失稳躲避蒙特卡洛树搜索算法(IAMCTS)用于容错步态规划.将容错步态规划问题转化为马尔科夫序列优化过程;根据稳定性判据制定了失稳状态躲避策略,并将其引入蒙特卡洛树搜索算法模拟阶段动作选择策略中,避免选择失稳支撑动作以生成容错步态序列.仿真系统环境中进行模拟对比分析表明,所提规划算法具有一定可行性和优势性;在 IAMCTS规划的自由容错步态模式下,单腿失效六足机器人在稀疏立足地形中的平均前进距离增长了 88.45 cm,平均前进速度提高了 18.91%.
针对全景多目标侦察遥感图像传输速度慢、传输质量差等问题,采用GTX高速串行收发技术、大容量高速缓存技术,设计了一种基于现场可编程门阵列(FPGA)的全景图像采集与高速传输系统,主要包括图像采集模块、GTX图像传输模块、数据缓存模块以及图像显示模块.GTX图像传输模块选用光纤作为传输介质,SFP+作为传输接口,传输协议为 Aurora 8B10B single lane 4 byte,数据缓存模块选用DDR3 SDRAM作为系统的高速缓存单元.实验测试结果表明,系统最高可采集图像分辨率为500×104 pixels,最高可实现10.312 5 Gbps高速视频图像的光纤无失真传输,误码率约为10-12,为大容量图像采集和传输提供了一种可行的方案.
传统配电网故障选线模型大多采用零序电流时频特征作为选线判据,单一信息域特征无法全面反映故障信息差异、适用范围存在局限性.为了提高模型复杂工况下选线准确率,提出一种基于卷积注意力机制优化双分支残差网络(CBAM-ResNet)和多域特征融合的配电网故障选线方法.首先,利用变分模态分解-希尔伯特变换和格拉姆角场将采集的零序电流信号分别映射为二维时频域和空间域图像,构建能够全面的反映故障信息的多域图像训练集;其次,通过 CBAM-ResNet网络深层次挖掘并融合多域特征信息,卷积注意力机制能对多域特征的重要性进行区分,加快网络训练速度,提高分类准确性;最后,将融合特征输入全连接层实现对配电网故障线路的选取.仿真结果表明,该方法相比传统选线方法具有更高的选线精度和噪声鲁棒性.
针对探地雷达图像人工检测速度慢,任务量重,长时间人工识别容易产生疲劳,传统图像手段检测效率低等问题,提出了一种基于改进的YOLOv7的探地雷达土石堤坝空洞检测算法.基于江西省九江市永修县和辽宁省盘锦市盘山县制作了一份当地土质的探地雷达空洞数据集,通过清洗,增强,反演增广了数据集,达到3 000张正负样本标注图像.其次在基础YOLOv7算法的基础上更换了可以自动调整网络的宽度、深度和分辨率的EfficientNet骨干网络;引入了全局注意力机制(GAM)来更好的理解输入数据中的结构,提高了土石方堤坝空洞的检测精度.实验表明,改进的YOLOv7算法在自建数据集上取得了精准度达到了80.23%,平均精度 mAp@0.5 达到了84.43%,较基础的YOLOv7算法上 mAp@0.5提升了7.72%.召回率提升了11.9%,充分展现了算法的可行性,为堤坝空洞隐患快速检测提供了良好的技术支持.
针对传统控制器在永磁直驱风力发电机中进行最大功率跟踪时响应速度慢、追踪精度低等问题,提出了一种基于改进指数趋近律和非线性扩张状态观测器(non-linear extended state observer,NLESO)的滑模控制方法.该方法使用连续函数代替符号函数 sign(s),引入增益函数设计改进指数趋近律以加快系统响应速度;将传统扩张状态观测器(extended state ob-server,ESO)中fal 函数替换成一种非线性函数,并基于此设计 NLESO 以提升系统抗扰动能力,结合二者设计改进滑模控制器取代传统PI控制器.在 MATLAB/Simulink上构建模型,结果表明,渐变风突变过程无超调,风速突变后再次平衡时间从0.054 s减少到 0.028 s,自然风过程中系统能精准追踪上额定转速,证明了该控制方法具有超调量小、响应速度快、追踪精度高等特点.
车道线检测技术是实现汽车无人驾驶的关键技术之一,能够帮助无人驾驶更快地实时处理从相机捕获的图像.近年来,研究人员在车道线检测的精度上取得了很大的进展,但在实际驾驶中,车辆可能会因有限的控制器算力而对车道线检测的速度和准确度产生影响.提出了一种动态识别感兴趣区域的方法,以Donkey car自动避障小车为主体进行实车试验,通过实时减少摄像头采集的图像中干扰信息,更加准确地识别车道线,在采集的图像数量不变的情况下,提高了卷积神经网络(convolutional neural networks,CNN)模型训练的速度和小车自动驾驶时车道线识别的准确度,进而减少了因车道线误判而存在的压线和驶离路面的情况.优化后的模型比原模型在训练时间方面缩短了约 38.71%,在小车自动驾驶时间上缩短了约21.67%,同时小车在转弯处行驶速度更加均匀,左右摇摆幅度减小.结果表明,所提方法具有良好的检测效果和准确率.
针对现有医学图像加密算法信息分散能力偏弱、安全性能偏低等问题,提出了一种基于双混沌结构、DNA 动态编码和分块扩散的医疗数位影像传输协定(digital imaging and communications in medicine,DICOM)医学图像加密算法.设计出一种新的三维超混沌系统并引入超混沌Chen系统构成双混沌迭代生成体系,避免单一系统遭到逆向解构分析从而被破解的风险.优化DNA编码机制,与混沌序列相结合,实现动态规则的DNA编码,提高算法的抗解码能力.此外,在扩散阶段对矩阵信息进行分块并采用类Zigzag式循环遍历法对像素信息逐位处理.经各项安全性分析实验,证明密文图像在统计特性、信息熵、密钥敏感性及空间大小等方面具备良好的加密性能并拥有较高的安全性,在医学图像加密领域中具备一定的前景.
针对线控转向用智能TAS传感器复杂性能的检测需求,研究分析了传感器的基本原理,利用 USBCAN 接口、STM32模块和伺服电机等硬件搭建了基于 LabVIEW的传感器测试系统.系统以调用动态链接库 TSMaster.dll 函数的方式进行上位机和下位机之间的串口通信,实现了对传感器运行工况的模拟,并利用智能汽车接口实现了智能 TAS传感器的协议解析,在 LabVIEW开发环境中采用通知器模式实现了传感器与伺服电机信号的同步采集,同时设计了离散傅里叶分析、转角游标算法等数据分析模块对采集到的信号进行处理.测试数据表明,传感器扭矩信号输出偏差始终保持在±0.15°范围内,方向盘转角误差在±0.7°范围内,所设计的测试系统能在理想范围内对智能TAS传感器进行扭矩和转角性能的复合式测量,加快了厂商研发以及投产的进度,具有实际应用价值.
为解决城市复杂地物环境中小面积森林检测的难题,提出了一种基于空间机器学习的城市小面积森林监测方法.该方法首先对卫星遥感大数据预处理,抑制云层和图像噪声.其次,提出了基于随机森林(random forest,RF)建立了森林识别模型.最后,选取了一个典型国家森林城市-江苏省南通市进行推广应用,实现了 10 m 空间分辨率城市森林变化监测.模型精度结果表明,对 2021 年南通市小面积森林覆盖精度为 99.25%.在此基础上,利用时间序列差分方法得到南通市 2019～2021 年森林变化图.通过对比验证结果表明,提出的方法可有效监测小面积森林,将对测准城市森林系统碳汇具有重要的参考意义.
针对 RRT*算法采样效率低、搜索时间长等问题,提出了一种基于分区思想的两阶段 RRT*算法(TP-RRT*).在探索阶段,利用自适应采样限制采样点的随机性,根据平行四边形法则合成新的矢量节点,既提升了采样点的方向性,又增强了算法对地图的搜索能力.在寻找最近点时,遍历整个随机树的过程存在计算量大、耗时长的问题,因此采用分区存储和检索的方法提升搜索的效率,同时限定分区中的节点数量,避免同一区域过度搜索.并在此基础上,生成一个新的父节点来改进原始算法中的重选父节点环节,进一步减少路径长度.最终在优化阶段时,通过对全局区域进行随机采样,并构造评估函数,得到可以改善路径的节点.仿真实验结果表明,改进算法与原算法在 3 种不同地图下,路径长度均减少了 10%以上、时间均缩短了 61%以上、节点数量均降低了 75%以上;TP-RRT*算法在路径规划的长度、时间、节点数等方面都优于 Q-RRT*和informed-RRT*.
为了提高地铁室内PM2.5 的预测精度,降低监测成本,提出了一种基于孤立森林算法(isolated forest,IF)、同步压缩小波变换算法(SWT)、改进麻雀搜索算法(improved sparrow search algorithm,ISSA)和长短期记忆网络(long short-term memory network,LSTM)的混合模型.首先,使用孤立森林算法检测并去除异常数据,在用SWT算法对原始PM2.5 数据进行去噪处理;其次,针对麻雀算法(SSA)易陷入局部最优、收敛速度慢的问题,利用正弦混沌、动态自适应惯性权重、高斯变异和反向学习策略改进麻雀算法,降低了SSA陷入局部最优解的概率,提高了麻雀算法的收敛速度和寻优能力;最后,利用ISSA对LSTM模型的参数进行寻优,构建ISSA-LSTM模型进行预测,得到最终的PM2.5 预测结果.实验结果表明,SWT-ISSA-LSTM模型在均方根误差比SWT-LSTM模型和SWT-SSA-LSTM模型分别降低了8.38和3.27 μg/m3.在拟合度方面,该模型比SWT-LSTM模型和SWT-SSA-LSTM模型分别高了10.6%和2.9%.
针对光伏电池电致发光(electroluminescence,EL)缺陷图像存在复杂背景干扰、缺陷目标长宽比和尺寸变化大等问题,在YOLOv5s网络的基础上提出一种改进的YOLOv5s-GD检测模型.首先在特征提取层融合GhostNetV2 模块,通过聚合本地和远程信息同步,在压缩模型参数量的同时增强模型对缺陷特征的提取能力;其次在特征融合层引入改进的DenseNet密集网络模块,通过特征重用进一步深度融合多级特征并改善信息传递,有效减少缺陷图像检测中漏检误检的情况;最后添加坐标注意力(CA)机制模块,兼顾通道和空间信息进行特征增强,提高了模型对缺陷目标区域的关注程度.与YOLOv5s原网络相比,YOLOv5s-GD检测模型在增长少量参数的情况下mAP@0.5 提升了3.3%,mAP@0.5:0.95 提升了 2.8%.研究结果表明,提出的YOLOv5s-GD检测模型对光伏电池EL缺陷图像可以有效地定位识别,检测精度有了显著的提高,同时检测速度达到75 fps,为实际工业生产环境中光伏电池EL缺陷图像检测提供技术参考.
高空带电作业机器人对目标物的精细化作业,需要合适的视觉辅助系统判断机械臂或爪具的位置与距离.提出一种将基于双目立体视觉的带电作业机器人目标识别与定位方法.通过对传统的Census算法进行改进,用窗口均值像素代替中心像素,用自适应窗口代替固定窗口,并与差的绝对和(SAD)算法加权融合,得到一种改进的SAD-Census立体匹配融合算法.利用该算法进行像素匹配运算,得出视差图;在YOLO框架下训练输电线螺栓数据集,获得螺栓深度距离并精准识别.螺栓定位实验结果表明,该方法在近距离定位时能达到1.2%的定位精度,最小相对误差为 0.5%,能够为带电作业机器人提供实时准确的环境感知信息,提高了带电作业机器人作业的稳定性.
