WO2022183698A1 - 基于输电线路暂态波形的人工智能故障辨识系统及方法 - Google Patents

基于输电线路暂态波形的人工智能故障辨识系统及方法 Download PDF

Info

Publication number
WO2022183698A1
WO2022183698A1 PCT/CN2021/115644 CN2021115644W WO2022183698A1 WO 2022183698 A1 WO2022183698 A1 WO 2022183698A1 CN 2021115644 W CN2021115644 W CN 2021115644W WO 2022183698 A1 WO2022183698 A1 WO 2022183698A1
Authority
WO
WIPO (PCT)
Prior art keywords
transmission line
transient waveform
deep learning
model
data
Prior art date
Application number
PCT/CN2021/115644
Other languages
English (en)
French (fr)
Inventor
王宇
李健
李哲
刘宇舜
白冰洁
周赞东
冯智慧
谢迎谱
雷梦飞
黎炎
韩冬
Original Assignee
国网电力科学研究院武汉南瑞有限责任公司
南瑞集团有限公司
国网安徽省电力有限公司
国网安徽省电力有限公司电力科学研究院
国家电网有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 国网电力科学研究院武汉南瑞有限责任公司, 南瑞集团有限公司, 国网安徽省电力有限公司, 国网安徽省电力有限公司电力科学研究院, 国家电网有限公司 filed Critical 国网电力科学研究院武汉南瑞有限责任公司
Publication of WO2022183698A1 publication Critical patent/WO2022183698A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/20Administration of product repair or maintenance
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01RMEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08Locating faults in cables, transmission lines, or networks
    • G01R31/081Locating faults in cables, transmission lines, or networks according to type of conductors
    • G01R31/085Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution lines, e.g. overhead
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01RMEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08Locating faults in cables, transmission lines, or networks
    • G01R31/088Aspects of digital computing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/04Inference or reasoning models
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06Energy or water supply
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00Aspects of pattern recognition specially adapted for signal processing
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04SSYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
    • Y04S10/52Outage or fault management, e.g. fault detection or location

Definitions

  • the invention relates to the field of power system fault identification, in particular to an artificial intelligence fault identification system and method based on the transient waveform of a transmission line.
  • the purpose of the present invention is to propose an artificial intelligence fault identification system and method based on the transient waveform of a transmission line, aiming to solve the above problems of difficulty in ensuring the timeliness of fault identification by human experts and inconsistent experience.
  • the present invention proposes an artificial intelligence fault identification method based on the transient waveform of a transmission line, which includes the following steps:
  • the real-time transmission line transient waveform sequence image data is input into the deep learning model, and the deep learning model performs feature extraction and forward reasoning of the sequence image data, and finally obtains the real-time transmission line transient waveform signal corresponding to each transmission line
  • the confidence level of the line fault type is used to obtain the fault type of the transmission line; the deep learning model is based on the sample database of the corresponding relationship between the historical transient waveform of the transmission line and the fault of the transmission line, and is obtained by using the deep learning framework.
  • the artificial intelligence fault identification system based on the transient waveform of the transmission line designed by the present invention includes a signal preprocessing module and a waveform identification module; the signal preprocessing module performs real-time transmission line transient waveform signal preprocessing based on the sliding window method, and obtains Corresponding real-time transmission line transient waveform sequence image data; the waveform identification module is used to input the real-time transmission line transient waveform sequence image data into the deep learning model, and the deep learning model performs feature extraction and analysis of the sequence image data. Through forward reasoning, the confidence of each transmission line fault type corresponding to the real-time transmission line transient waveform signal is finally obtained, and the transmission line fault type is obtained.
  • a recognition model for automatically identifying the transient waveform of a transmission line can be established, and line faults caused by lightning strikes or non-lightning strikes can be identified and classified in a timely, efficient and accurate manner, replacing human experts and reducing human expert labor. At the same time overcome the possible problems of human experts.
  • the identification model constructed by the present invention can be evolved by training more sample data, which is beneficial to continuously improve the accuracy of fault identification of the model.
  • the model trained according to the present invention can be deployed in an embedded hardware device and run well, can realize automatic identification of faults at the device end, reduce the communication transmission of a large number of waveforms, and improve the real-time performance of information feedback.
  • Fig. 1 is the system structure schematic diagram of the present invention
  • Fig. 2 is the flow chart of the method of the present invention.
  • Fig. 3 is the neural network structure diagram of the deep learning model of the present invention.
  • Fig. 4 is the color image generated after the original image of the waveform signal of the positive example and the GADF/GASF image processing of the example of the present invention
  • Fig. 5 is the color image that the original image of the counter example waveform signal and the GADF/GASF image processing of the example of the present invention are generated;
  • Fig. 6 is the lightning strike waveform prediction case of the example of the present invention.
  • Fig. 7 is the non-lightning waveform prediction case of the example of the present invention.
  • 1-signal preprocessing module 2-manual labeling module, 3-waveform recognition module, 4-training and tuning module, 5-performance testing and optimization module.
  • An artificial intelligence fault identification method based on the transient waveform of a transmission line proposed by the present invention, as shown in Figure 2, includes the following steps:
  • the real-time transmission line transient waveform sequence image data is input into the deep learning model, and the deep learning model performs feature extraction and forward reasoning of the sequence image data, and finally obtains the real-time transmission line transient waveform signal corresponding to each transmission line
  • the confidence level of the line fault type is used to obtain the transmission line fault type
  • the deep learning model is based on the sample database of the corresponding relationship between the historical transmission line transient waveform and the transmission line fault, and is obtained by using the deep learning framework;
  • the method for establishing the deep learning model specifically includes the following steps:
  • step S23 using the historical transient waveform sequence image data and the sample library output in step S21 to establish a deep learning model by using the Pytorch deep learning framework.
  • the transmission line transient waveform signal is induced by the transmission line fault, and the waveform signals caused by different faults have different characteristics, and the corresponding transmission line transient waveform signal and fault type are manually marked to form a data label.
  • waveform signal normalization unify the duration and sampling frequency of the transient waveform signal of the transmission line, the unified sampling rate is 2000000HZ, and the unified sampling point number is 2400;
  • waveform signal segmentation the sampling length control method based on the sliding window divides the transmission line transient waveform into several overlapping parts, and obtains the transmission line transient waveform sequence data;
  • waveform signal transformation transform the transient waveform sequence data of the transmission line based on the GAF (Gramian Angular Field) method to obtain the corresponding sequence image data.
  • the normalized transmission line transient waveform signal is divided into several overlapping parts, and the size of the overlapping area is half the size of the sliding window, that is, the sliding window is moved forward by half a window unit each time, and the window is slid.
  • the obtained sequence data generates transmission line transient waveform signal sequence data.
  • the specific method of the waveform signal conversion in step S22.3 is:
  • GAF matrix uses trigonometric functions to obtain GAF matrix, including GASF (Gramian Angular Summation Field) and GADF (Gramian Angular Difference Field);
  • the Gramian angle difference field GADF is obtained by using the cos function of the difference between the two angles.
  • the calculation method of the GADF is as follows:
  • step S3 the specific method for performing parameter training and tuning of the deep learning model in step S3 is:
  • S31 data set division: divide the data set into training set and test set according to 7:3;
  • optimizer settings select Adam (Adaptive Moment Estimation) optimizer;
  • model parameter training and tuning use the training set divided in step S31 for model training, complete iterative B times on the training set, observe the convergence of the model, and use the early stopping strategy to prevent the model from seriously
  • the phenomenon of overfitting that is, stop the iteration before the model converges on the training data set to prevent overfitting: take the model parameter file with the highest accuracy on the test set and save it. After the training is completed, the optimal model will be obtained. parameter file;
  • the calculation method of the accuracy rate is:
  • TP True Positive: represents the actual positive example, and the model judges it as a positive example
  • FP False Positive: represents the actual positive example, and the model judges it as a negative example;
  • TN True Negative: represents the actual negative example, and the model judges it as a negative example
  • the specific method for performing the performance test of the deep learning model and the targeted calculation optimization in the step S4 is:
  • the deep learning model is shown in Figure 3, and the input of the deep learning model network is nine frame sequence pictures, all of which are three-channel color pictures, and the size is 112 ⁇ 112 ⁇ 3 (here represents three-dimensional data, The ' ⁇ ' symbol is used to distinguish different dimensions, the same below), the output of the network is the recognition result, and the result contains the normalized probability value of each category; in the network structure, the input sequence picture will first pass through the 3D volume Product and 3D pooling part, which extracts temporal features from this part of the network, where the size of the 3D convolution kernel is 5 ⁇ 7 ⁇ 7, the size of the convolution stride is 1 ⁇ 2 ⁇ 2, and the size of the padding is 2 ⁇ 3 ⁇ 3, the sliding window size of the 3D pooling part is 1 ⁇ 3 ⁇ 3, the stride size is 1 ⁇ 2 ⁇ 2, and the padding size is 0 ⁇ 1 ⁇ 1.
  • the 34-layer Resnet is used as the backbone network (Backbone) to extract spatial semantic features, and finally the 1D convolution layer and the continuous fully connected layer complete the decoding (Decode), and the normalized probability value of each category is output through the Softmax layer ( The value range is [0, 1]), and all the forward reasoning process is completed.
  • the model construction, parameter training and tuning, and performance optimization all use the Pytorch deep learning framework, which is used in conjunction with the GPU (model: 2080 Ti 11G) issued by NVIDIA.
  • the accuracy of the model can reach 99.8% on the training set and 92.6% on the test set; on the PC side (hardware model: CPU: i7 8700, GPU: GTX 1070 8G) graphics processing
  • the time is 0.04s, and the recognition processing time is 0.02s
  • the embedded side NVIDIA Jetson Nano
  • the graphics processing time is 0.05s
  • the recognition processing time is 0.15s.
  • Figures 4 and 5 are images after GADF/GASF imaging.
  • the three sub-images from left to right in Figure 3 are the original waveform, GADF image and GASF image of the positive example (lightning waveform) respectively; the three sub-images from left to right in Figure 4 are the original waveform of the negative example (non-lightning waveform) respectively Waveform, GADF image, GASF image.
  • Figures 6 and 7 are examples of the recognition results of the deep learning model.
  • Fig. 6 is a case of lightning strike waveform prediction
  • Fig. 7 is an example case of non-lightning strike waveform prediction.
  • the top text gt represents the ground truth, and pred represents the prediction of the deep learning model.
  • ' symbol in the score are category one (lightning strike) and category two ( non-lightning strikes) confidence.
  • An artificial intelligence fault identification system based on the transient waveform of transmission line includes signal preprocessing module 1, manual labeling module 2, waveform identification module 3, training and tuning module 4 and performance testing and optimization Module 5; the signal preprocessing module 1 completes the historical transmission line transient waveform signal preprocessing based on the sliding window method, and obtains the corresponding historical transient waveform sequence image data; the manual labeling module 2 is aimed at historical transmission line transients The waveform signal is manually marked with the fault type, the corresponding fault type data label is generated, and the sample library of the corresponding relationship between the transient waveform of the transmission line and the fault of the transmission line is established by using all the fault type data labels; the waveform identification module 3 uses the historical transient waveform sequence.
  • the image data and the sample library output in step S21 are used to establish a deep learning model using the Pytorch deep learning framework; the signal preprocessing module 1 performs real-time transmission line transient waveform signal preprocessing based on the sliding window method to obtain corresponding real-time transmission lines.
  • Transient waveform sequence image data the waveform identification module 3 inputs the real-time transmission line transient waveform sequence image data into the deep learning model, and the deep learning model performs feature extraction and forward reasoning of the sequence image data, and finally obtains real-time The confidence level of each transmission line fault type corresponding to the transmission line transient waveform signal is obtained, and the transmission line fault type is obtained; the training and tuning module 4 is used to complete the parameter training and tuning of the deep learning model; the performance test and The optimization module 5 is used to complete the performance test and targeted calculation optimization of the deep learning model.
  • the sequence image data output end of the signal preprocessing module 1 is connected to the input end of the waveform identification module 3, and the waveform classification label data of the manual labeling module 2 is connected to the input end of the signal preprocessing module 1.
  • terminal, the waveform classification label data output terminal of the manual labeling module 2 is connected to the input terminal of the training and tuning module 4
  • the model output terminal of the waveform identification module 3 is connected to the input terminal of the training and tuning module 4
  • the output terminal of the training and tuning module 4 is connected to the input terminal of the performance testing and optimization module 5 .

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Human Resources & Organizations (AREA)
  • Artificial Intelligence (AREA)
  • Economics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Strategic Management (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Image Analysis (AREA)

Abstract

一种输电线路暂态波形的人工智能故障辨识方法,主要包括以下步骤:针对原始波形信号进行故障类型人工标注生成数据标签,建立输电线路暂态波形故障样本库;基于滑动窗口方法完成输电线路暂态波形信号预处理,获得对应的序列图像数据;搭建深度学习模型,实现输电线路暂态波形识别;对深度学习模型进行参数训练与调优;对深度学习模型进行性能测试,并完成针对性优化以提升性能,以实现快速、精确、可靠的输电线路暂态波形辨识。此外,还公开了实现方法的输电线路暂态波形的人工智能故障辨识系统,包括信号预处理模块、人工标注模块、波形识别模块、训练与调优模块和性能测试与优化模块。

Description

基于输电线路暂态波形的人工智能故障辨识系统及方法 技术领域
本发明涉及电力系统故障识别领域,尤其涉及一种基于输电线路暂态波形的人工智能故障辨识系统及方法。
背景技术
输电线路途经的地域条件复杂多样,极易遭受雷电、污秽、动植物、大风、覆冰等各种自然因素的影响而发生跳闸事故。线路跳闸对系统安全、设备安全以及供电可靠性都有较大影响,我国电网运行规程要求在跳闸后最短的时间内查明故障位置及故障类型。因此,快速、精确、可靠的故障辨识和定位技术对输电线路安全运行尤为重要。
目前,我国输电线路安装了分布式故障行波检测装置,可以测量线路上的暂态电流波形。根据记录到的行波电流,人类专家可辨别雷击和非雷击故障(主要包括污秽闪络、树枝闪络、冰闪、鸟闪、大风舞动等),雷击故障又可具体区分绕击和反击故障。中国专利CN102788923A提出了一种输电线路雷电绕击故障的辨识方法,主要利用行波电流幅值最大值前的一小段同极性脉冲来辨识绕击故障。实际故障处置中,专家通过综合研判行波波形形状和特征参数得到辨识结果。客观上讲,正确辨识线路故障要求人类专家专业知识和经验丰富。人类专家辨识故障主要问题在于及时性不够、经验不容易复制以及不同专家识别结果存在差异。
发明内容
本发明的目的就是要提出一种基于输电线路暂态波形的人工智能故障辨识系统及方法,旨在解决上述人类专家辨识故障及时性难以保证、经验不统一等问题。
为实现此目的,本发明提出了一种基于输电线路暂态波形的人工智能故障辨识方法,包括如下步骤:
S1,基于滑动窗口方法进行实时的输电线路暂态波形信号预处理,获得对应的实时的输电线路暂态波形序列图像数据;
S2,将实时的输电线路暂态波形序列图像数据输入到深度学习模型中,由深度学习模型进行序列图像数据的特征提取和前向推理,最终获得实时的输电线路暂态波形信号对应的各个输电线路故障类型的置信度,得到输电线路故障类型;所述深度学习模型是基于历史的输电线路暂态波形与输电线路故障对应关系的样本库利用深度学习框架建立得到。
本发明设计的基于输电线路暂态波形的人工智能故障辨识系统,包括信号预处理模块和波形识别模块;所述信号预处理模块基于滑动窗口方法进行实时的输电线路暂态波形信号预处理,获得对应的实时的输电线路暂态波形序列图像数据;所述波形识别模块用于将实时的输电线路暂态波形序列图像数据输入到深度学习模型中,由深度学习模型进行序列图像数据的特征提取和前向推理,最终获得实时的输电线路暂态波形信号对应的各个输电线路故障类型的置信度,得到输电线路故障类型。
本发明的有益效果:
1、通过本发明可以建立机器自动辨识输电线路暂态波形的识别模型,可以及时、高效、准确的对由雷击或非雷击引起的线路故障进行辨识和分类,替代人类专家,减少人类专家劳力,同时克服人类专家可能存在的问题。
2、通过本发明构建的识别模型可以通过训练更多样本数据得到进化,有利于持续提高模型辨识故障准确度。
3、按照本发明训练的模型可以部署到嵌入式硬件装置中良好运行,可以实现装置端的故障自动识别,减少大量波形的通讯传输,提高信息反馈的实时性。
附图说明
图1为本发明的系统结构示意图;
图2为本发明方法流程图;
图3为本发明深度学习模型的神经网络结构图;
图4为本发明实例的对正例波形信号原始图像以及GADF/GASF图像化处理后生成的彩色图像;
图5为本发明实例的对反例波形信号原始图像以及GADF/GASF图像化处理后 生成的彩色图像;
图6为本发明实例的雷击波形预测案例;
图7为本发明实例的非雷击波形预测案例;
其中,1-信号预处理模块、2-人工标注模块、3-波形识别模块、4-训练与调优模块、5-性能测试与优化模块。
具体实施方式
以下结合附图和具体实施例对本发明作进一步的详细说明:
本发明所提出的一种基于输电线路暂态波形的人工智能故障辨识方法,如图2所示,它包括如下步骤:
S1,基于滑动窗口方法进行实时的输电线路暂态波形信号预处理,获得对应的实时的输电线路暂态波形序列图像数据;
S2,将实时的输电线路暂态波形序列图像数据输入到深度学习模型中,由深度学习模型进行序列图像数据的特征提取和前向推理,最终获得实时的输电线路暂态波形信号对应的各个输电线路故障类型的置信度,得到输电线路故障类型;所述深度学习模型是基于历史的输电线路暂态波形与输电线路故障对应关系的样本库利用深度学习框架建立得到;
S3,进行深度学习模型的参数训练与调优;
S4,进行深度学习模型的性能测试和针对性计算优化。
上述技术方案中,所述深度学习模型的建立方法具体包括如下步骤:
S21,针对历史的输电线路暂态波形信号进行故障类型人工标注,生成对应的故障类型数据标签,利用所有故障类型数据标签建立输电线路暂态波形与输电线路故障对应关系的样本库;
S22,基于滑动窗口方法完成历史的输电线路暂态波形信号预处理,获得对应的历史暂态波形序列图像数据;
S23,利用历史暂态波形序列图像数据与步骤S21输出的样本库,使用Pytorch深度学习框架建立深度学习模型。
所述步骤S1中输电线路暂态波形信号由输电线路故障感应引起,不同故障引 起的波形信号具有不同的特征,针对相应的输电线路暂态波形信号和故障类型进行人工标注形成数据标签。
上述技术方案中,所述步骤S22完成输电线路暂态波形信号预处理的具体方法为:
S22.1,波形信号归一化:统一输电线路暂态波形信号的持续时间和采样频率,统一采样率为2000000HZ,统一采样点数为2400;
S22.2,波形信号切分:基于滑动窗口的采样长度控制方法将输电线路暂态波形切分为有重叠的若干份,获得输电线路暂态波形序列数据;
S22.3,波形信号转化:基于GAF(Gramian Angular Field,格拉米安角度场)方法对输电线路暂态波形序列数据进行转化,得到对应的序列图像数据。
上述技术方案中,所述步骤S22.2波形信号切分的具体方法为:
将归一化后的输电线路暂态波形信号切分为有重叠的若干份,重叠区域的大小为滑动窗口大小的一半,即每次将滑动窗口向前移动半个窗口单位,并将窗口滑动所获得的序列数据生成输电线路暂态波形信号序列数据。
上述技术方案中,所述步骤S22.3波形信号转化的具体方法为:
S22.3.1,将输电线路暂态波形信号序列数据转化为极坐标系数据格式;
输电线路暂态波形信号的时间序列是X={x 1,x 2,...,x n},其长度为n,使用公式(1)将输电线路暂态波形信号时间序列归一化缩放到[-1,1]范围内:
Figure PCTCN2021115644-appb-000001
式中
Figure PCTCN2021115644-appb-000002
使用三角函数来代替归一化之后的值,用
Figure PCTCN2021115644-appb-000003
来表示使用公式(1)归一化之后的时间序列,令
Figure PCTCN2021115644-appb-000004
因此,
Figure PCTCN2021115644-appb-000005
S22.3.2,使用三角函数获得GAF矩阵,包含GASF(Gramian Angular Summation Field,格拉米安角求和场)和GADF(Gramian Angular Difference Field,格拉米安角差分场)两种;
将公式(1)中归一化的数据,使用两角和的cos函数获得格拉米安角求和场GASF:
定义矩阵GASF为
Figure PCTCN2021115644-appb-000006
于是,有
Figure PCTCN2021115644-appb-000007
Figure PCTCN2021115644-appb-000008
T表示转置,可以得到:
Figure PCTCN2021115644-appb-000009
同理,使用两角差的cos函数获得格拉米安角差分场GADF,计算GADF方式如下:
Figure PCTCN2021115644-appb-000010
S22.3.3,生成三通道彩色图像:
对获得的大小为n×n的GADF和GASF矩阵数据,使用matplotlib库中的pyplot应用函数接口(API),生成对应的尺寸为n×n×3的三通道彩色图像。
上述技术方案中,所述步骤S3进行深度学习模型的参数训练与调优的具体方法为:
S31,数据集划分:将数据集按7:3分为训练集和测试集;
S32,优化器设置:选择Adam(Adaptive Moment Estimation)优化器;
S33,设置学习率初始值及调整策略:设置学习率(Learning Rate)初始值为0.01,配合余弦退火(Cosine Decay)学习率动态策略调整学习率的大小,使学习率按照周期变化:即在一个完整的训练周期内,学习率大小按照余弦函数变化规律从初始值0.01逐渐降到0,再逐渐升回0.01;
S34,模型参数训练及调优:使用步骤S31划分好的训练集进行模型训练,在训练集上完整迭代B余次,观察模型收敛情况,并使用早停(Early stopping)策略以防止模型发生严重的过拟合(over fitting)现象,即在模型对训练数据集迭代 收敛之前停止迭代来防止过拟合:取测试集上准确率最高的模型参数文件并保存,训练完成后即获得最优的参数文件;
所述准确率的计算方法为:
Figure PCTCN2021115644-appb-000011
式(5)中所用缩写分别为:
TP(True Positive):代表实际为正例,且模型判断为正例;
FP(False Positive):代表实际为正例,且模型判断为反例;
TN(True Negative):代表实际为反例,且模型判断为反例;
FN(False Negative):代表实际为反例,且模型判断为正例。
上述技术方案中,所述步骤S4进行深度学习模型的性能测试和针对性计算优化的具体方法为:
S41,在提高模型推理速度上,对同类型计算任务进行任务合并和多线程处理,可节省60%的计算耗时;
S42,在改善模型识别精度上,针对数据集中模型识别的错例进行误差分析,并针对该数据进行微调。
上述技术方案中,所述深度学习模型如图3所示,深度学习模型网络的输入为九帧序列图片,均为三通道的彩色图片,尺寸为112×112×3(此处表示三维数据,‘×’符号用于区分不用的维度,下同),网络的输出为识别的结果,结果中包含每个类别的归一化概率值;在网络结构中,输入的序列图片会首先通过3D卷积和3D池化部分,由此部分网络提取时域特征,其中3D卷积核大小为5×7×7,卷积步长(Stride)大小为1×2×2,填充(Padding)大小为2×3×3,3D池化部分滑动窗口的大小为1×3×3,步长大小为1×2×2,填充大小为0×1×1。此后通过34层的Resnet作为骨干网络(Backbone)来提取空间语义特征,最后由1D卷积层和连续的全连接层完成解码(Decode),通过Softmax层输出每个类别的归一化概率值(取值范围[0,1]),至此完成全部的前向推理过程。
上述技术方案中,所述模型搭建、参数训练与调优和性能优化均采用Pytorch深度学习框架,配合英伟达(NVIDIA)公司发行的GPU(型号:2080 Ti 11G)使用。 至于性能评测方面,该模型在训练集上准确率可以达到99.8%,同时测试集上也有92.6%的准确率;在PC端(硬件型号为CPU:i7 8700,GPU:GTX 1070 8G)图形化处理时间为0.04s,识别处理时间为0.02s;而在嵌入式端(NVIDIA Jetson Nano)图形化处理时间为0.05s,识别处理时间为0.15s。
对于不同参数设置下的准确率及推理耗时如下表所示:
Figure PCTCN2021115644-appb-000012
图4和图5为GADF/GASF图像化后的图像。其中,图3中从左到右三个子图分别为正例(雷击波形)的原始波形、GADF图像、GASF图像;图4中从左到右三个子图分别为反例(非雷击波形)的原始波形、GADF图像、GASF图像。图6和图7为深度学习模型识别结果的案例。其中,图6为雷击波形预测案例,图7为非雷击波形预测举例案例。图示中顶端文字gt代表真实的类别(ground truth),pred表示深度学习模型预测的类别(prediction),score中用’|’符号间隔的两个数值分别为类别一(雷击)和类别二(非雷击)的置信度。
一种基于输电线路暂态波形的人工智能故障辨识系统,如图1所示,它包括信号预处理模块1、人工标注模块2、波形识别模块3、训练与调优模块4和性能测试与优化模块5;所述信号预处理模块1基于滑动窗口方法完成历史的输电线路暂态波形信号预处理,获得对应的历史暂态波形序列图像数据;所述人工标注模块2针对历史的输电线路暂态波形信号进行故障类型人工标注,生成对应的故障类型数据标签,利用所有故障类型数据标签建立输电线路暂态波形与输电线路故障对应关系的样本库;所述波形识别模块3利用历史暂态波形序列图像数据与步骤S21输 出的样本库,使用Pytorch深度学习框架建立深度学习模型;所述信号预处理模块1基于滑动窗口方法进行实时的输电线路暂态波形信号预处理,获得对应的实时的输电线路暂态波形序列图像数据;所述波形识别模块3将实时的输电线路暂态波形序列图像数据输入到深度学习模型中,由深度学习模型进行序列图像数据的特征提取和前向推理,最终获得实时的输电线路暂态波形信号对应的各个输电线路故障类型的置信度,得到输电线路故障类型;所述训练与调优模块4用于完成深度学习模型的参数训练与调优;所述性能测试与优化模块5用于完成深度学习模型的性能测试和针对性计算优化。
上述技术方案中,所述信号预处理模块1的序列图像数据输出端连接所述波形识别模块3的输入端,所述人工标注模块2的波形分类标签数据连接所述信号预处理模块1的输入端,所述人工标注模块2的波形分类标签数据输出端连接所述训练与调优模块4的输入端,所述波形识别模块3的模型输出端连接所述训练与调优模块4的输入端,所述训练与调优模块4的输出端连接所述性能测试与优化模块5的输入端。
本说明书未作详细描述的内容属于本领域专业技术人员公知的现有技术。

Claims (9)

  1. 一种基于输电线路暂态波形的人工智能故障辨识方法,其特征在于,它包括如下步骤:
    S1,基于滑动窗口方法进行实时的输电线路暂态波形信号预处理,获得对应的实时的输电线路暂态波形序列图像数据;
    S2,将实时的输电线路暂态波形序列图像数据输入到深度学习模型中,由深度学习模型进行序列图像数据的特征提取和前向推理,最终获得实时的输电线路暂态波形信号对应的各个输电线路故障类型的置信度,得到输电线路故障类型;所述深度学习模型是基于历史的输电线路暂态波形与输电线路故障对应关系的样本库利用深度学习框架建立得到。
  2. 根据权利要求1所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于,所述深度学习模型的建立方法具体包括如下步骤:
    S21,针对历史的输电线路暂态波形信号进行故障类型人工标注,生成对应的故障类型数据标签,利用所有故障类型数据标签建立输电线路暂态波形与输电线路故障对应关系的样本库;
    S22,基于滑动窗口方法完成历史的输电线路暂态波形信号预处理,获得对应的历史暂态波形序列图像数据;
    S23,利用历史暂态波形序列图像数据与步骤S21输出的样本库,使用Pytorch深度学习框架建立深度学习模型。
  3. 根据权利要求1所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于,还包括如下步骤:
    S3,进行深度学习模型的参数训练与调优;
    S4,进行深度学习模型的性能测试和针对性计算优化。
  4. 根据权利要求2所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于:所述步骤S22完成输电线路暂态波形信号预处理的具体方法为:
    S22.1,将输电线路暂态波形信号的持续时间和采样频率设置为相同的值;
    S22.2,基于滑动窗口的采样长度控制方法将输电线路暂态波形切分为有重叠的若干份,获得输电线路暂态波形序列数据;
    S22.3,基于格拉米安角度场GAF方法对输电线路暂态波形序列数据进行转化,得到对应的序列图像数据。
  5. 根据权利要求4所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于:所述步骤S22.2的具体方法为:
    将归一化后的输电线路暂态波形信号切分为有重叠的若干份,重叠区域的大小为滑动窗口大小的一半,所获得的序列数据生成输电线路暂态波形信号序列数据。
  6. 根据权利要求4所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于:所述步骤S22.3的具体方法为:
    S22.3.1,将输电线路暂态波形信号序列数据转化为极坐标系数据格式;输电线路暂态波形信号的时间序列是X={x 1,x 2,...,x n},其长度为n,使用公式(1)将输电线路暂态波形信号时间序列归一化缩放到[-1,1]范围内:
    Figure PCTCN2021115644-appb-100001
    式中
    Figure PCTCN2021115644-appb-100002
    使用三角函数来代替归一化之后的值,用
    Figure PCTCN2021115644-appb-100003
    来表示使用公式(1)归一化之后的时间序列,令
    Figure PCTCN2021115644-appb-100004
    因此,
    Figure PCTCN2021115644-appb-100005
    S22.3.2,使用三角函数获得GAF矩阵,包含格拉米安角求和场GASF和格拉米安角差分场GADF两种;
    将公式(1)中归一化的数据,使用两角和的cos函数获得格拉米安角求和场GASF:
    定义矩阵GASF为
    Figure PCTCN2021115644-appb-100006
    于是,有
    Figure PCTCN2021115644-appb-100007
    Figure PCTCN2021115644-appb-100008
    T表示转置,可以得到:
    Figure PCTCN2021115644-appb-100009
    同理,使用两角差的cos函数获得格拉米安角差分场GADF,计算GADF方式如下:
    Figure PCTCN2021115644-appb-100010
    S22.3.3,对获得的大小为n×n的GADF和GASF矩阵数据,使用matplotlib库中的pyplot应用函数接口,生成对应的尺寸为n×n×3的三通道彩色图像。
  7. 根据权利要求3所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于:所述步骤S3进行深度学习模型的参数训练与调优的具体方法为:
    S31,将数据集按7:3分为训练集和测试集;
    S32,选择Adam优化器;
    S33,设置学习率初始值为A,配合余弦退火学习率动态策略调整学习率的大小,使学习率按照周期变化,在一个完整的训练周期内,学习率大小按照余弦函数变化规律从初始值A逐渐降到0,再逐渐升回A;
    S34,使用步骤S31划分好的训练集进行模型训练,在训练集上完整迭代B余次,观察模型收敛情况,并使用早停策略以防止模型发生严重的过拟合现象,取测试集上准确率最高的模型参数文件并保存,训练完成后获得最优的参数文件;
    所述准确率的计算方法为:
    Figure PCTCN2021115644-appb-100011
    式(5)中所用缩写分别为:
    TP:代表实际为正例,且模型判断为正例;
    FP:代表实际为正例,且模型判断为反例;
    TN:代表实际为反例,且模型判断为反例;
    FN:代表实际为反例,且模型判断为正例。
  8. 根据权利要求3所述的基于输电线路暂态波形的人工智能故障辨识方法,其特征在于:所述步骤S4进行深度学习模型的性能测试和针对性计算优化的具体 方法为:
    S41,在提高模型推理速度上,对同类型计算任务进行任务合并和多线程处理;
    S42,在改善模型识别精度上,针对数据集中模型识别的错例进行误差分析,并针对该数据进行微调。
  9. 一种基于输电线路暂态波形的人工智能故障辨识系统,其特征在于:包括信号预处理模块(1)、波形识别模块(3);
    所述信号预处理模块(1)基于滑动窗口方法进行实时的输电线路暂态波形信号预处理,获得对应的实时的输电线路暂态波形序列图像数据;
    所述波形识别模块(3)用于将实时的输电线路暂态波形序列图像数据输入到深度学习模型中,由深度学习模型进行序列图像数据的特征提取和前向推理,最终获得实时的输电线路暂态波形信号对应的各个输电线路故障类型的置信度,得到输电线路故障类型。
PCT/CN2021/115644 2021-03-05 2021-08-31 基于输电线路暂态波形的人工智能故障辨识系统及方法 WO2022183698A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110245717.2 2021-03-05
CN202110245717.2A CN113033837A (zh) 2021-03-05 2021-03-05 基于输电线路暂态波形的人工智能故障辨识系统及方法

Publications (1)

Publication Number Publication Date
WO2022183698A1 true WO2022183698A1 (zh) 2022-09-09

Family

ID=76468056

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/115644 WO2022183698A1 (zh) 2021-03-05 2021-08-31 基于输电线路暂态波形的人工智能故障辨识系统及方法

Country Status (2)

Country Link
CN (1) CN113033837A (zh)
WO (1) WO2022183698A1 (zh)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115494350A (zh) * 2022-11-21 2022-12-20 昆明理工大学 一种交流输电线路雷击故障识别方法及系统
CN115876475A (zh) * 2023-02-06 2023-03-31 山东大学 一种故障诊断方法、系统、设备及存储介质
CN116008731A (zh) * 2023-02-15 2023-04-25 重庆大学 配电网高阻故障辨识方法、装置和电子设备
CN116008756A (zh) * 2023-03-15 2023-04-25 国网福建省电力有限公司 电容式电压互感器绝缘故障诊断方法、系统、设备和介质
CN116087692A (zh) * 2023-04-12 2023-05-09 国网四川省电力公司电力科学研究院 一种配电网树线放电故障识别方法、系统、终端及介质
CN116827764A (zh) * 2023-08-23 2023-09-29 山西绿柳科技有限公司 一种基于神经网络的物联网故障检测控制方法及系统
CN117216485A (zh) * 2023-11-09 2023-12-12 国网山东省电力公司电力科学研究院 一种基于客观赋权的输电录波鸟害故障判别方法及系统
CN117250439A (zh) * 2023-11-08 2023-12-19 国网四川省电力公司电力科学研究院 一种面向多源接地故障的三层式研判分析系统
CN117310508A (zh) * 2023-11-30 2023-12-29 山东科技大学 一种快速准确测量锂电池电变量的方法
CN117313018A (zh) * 2023-11-29 2023-12-29 国网浙江省电力有限公司 一种电力输电线路异常状态识别方法及系统
CN117368639A (zh) * 2023-10-18 2024-01-09 浙江德清迪生电力科技有限公司 基于人工智能的输电线路故障识别系统及方法
CN117851868A (zh) * 2024-01-15 2024-04-09 珠海安瑞通电子科技有限公司 基于深度学习的多模型推进式配电网故障识别方法及系统

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113033837A (zh) * 2021-03-05 2021-06-25 国网电力科学研究院武汉南瑞有限责任公司 基于输电线路暂态波形的人工智能故障辨识系统及方法
CN113627547B (zh) * 2021-08-16 2024-01-26 河北工业大学 训练方法、电弧检测方法、装置、电子设备及存储介质
CN113834998B (zh) * 2021-08-28 2023-05-30 西安交通大学 一种基于电力线载波技术的飞机电缆故障检测识别方法
CN114252739B (zh) * 2021-12-24 2023-11-03 国家电网有限公司 配电网单相接地故障判别方法、系统、设备和存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3460494A1 (en) * 2017-09-26 2019-03-27 Siemens Aktiengesellschaft A method and apparatus for automatic detection of a fault type
CN109784276A (zh) * 2019-01-16 2019-05-21 东南大学 一种基于dbn的电压暂降特征提取与暂降源辨识方法
CN111553112A (zh) * 2020-03-16 2020-08-18 广西电网有限责任公司电力科学研究院 一种基于深度置信网络的电力系统故障辨识方法及装置
CN111611924A (zh) * 2020-05-21 2020-09-01 东北林业大学 一种基于深度迁移学习模型的蘑菇识别方法
CN111898729A (zh) * 2020-06-10 2020-11-06 国网江苏省电力有限公司电力科学研究院 一种输电线路故障原因识别方法及系统
CN113033837A (zh) * 2021-03-05 2021-06-25 国网电力科学研究院武汉南瑞有限责任公司 基于输电线路暂态波形的人工智能故障辨识系统及方法

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102788932B (zh) * 2012-06-13 2016-04-06 武汉三相电力科技有限公司 一种输电线路雷电绕击故障的辨识方法
CN106874950A (zh) * 2017-02-13 2017-06-20 云南电网有限责任公司电力科学研究院 一种暂态电能质量录波数据的识别分类方法
CN108959732B (zh) * 2018-06-15 2019-09-27 西安科技大学 一种基于卷积神经网络的输电线路故障类型识别方法
CN109324266B (zh) * 2018-11-21 2021-06-22 国网电力科学研究院武汉南瑞有限责任公司 一种基于深度学习的配网接地故障分析方法
CN110245617B (zh) * 2019-06-17 2022-06-24 江苏云脑数据科技有限公司 基于暂态录波波形的人工智能分析方法
CN110247420B (zh) * 2019-07-17 2020-07-28 四川轻化工大学 一种hvdc输电线路故障智能识别方法
CN110297146B (zh) * 2019-07-30 2020-08-04 华北电力大学 基于暂态波形特征的输电线路雷击干扰与故障识别方法
CN110907755A (zh) * 2019-12-03 2020-03-24 广西电网有限责任公司电力科学研究院 一种输电线路在线监测故障识别方法
CN111339872A (zh) * 2020-02-18 2020-06-26 国网信通亿力科技有限责任公司 一种基于分类模型的电网故障分类方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3460494A1 (en) * 2017-09-26 2019-03-27 Siemens Aktiengesellschaft A method and apparatus for automatic detection of a fault type
CN109784276A (zh) * 2019-01-16 2019-05-21 东南大学 一种基于dbn的电压暂降特征提取与暂降源辨识方法
CN111553112A (zh) * 2020-03-16 2020-08-18 广西电网有限责任公司电力科学研究院 一种基于深度置信网络的电力系统故障辨识方法及装置
CN111611924A (zh) * 2020-05-21 2020-09-01 东北林业大学 一种基于深度迁移学习模型的蘑菇识别方法
CN111898729A (zh) * 2020-06-10 2020-11-06 国网江苏省电力有限公司电力科学研究院 一种输电线路故障原因识别方法及系统
CN113033837A (zh) * 2021-03-05 2021-06-25 国网电力科学研究院武汉南瑞有限责任公司 基于输电线路暂态波形的人工智能故障辨识系统及方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
MATINDIFE LISTON, SUN YANXIA, WANG ZENGHUI: "Image-based mains signal disaggregation and load recognition", COMPLEX & INTELLIGENT SYSTEMS, vol. 7, no. 2, 1 April 2021 (2021-04-01), pages 901 - 927, XP055963553, ISSN: 2199-4536, DOI: 10.1007/s40747-020-00254-0 *

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115494350B (zh) * 2022-11-21 2023-03-24 昆明理工大学 一种交流输电线路雷击故障识别方法及系统
CN115494350A (zh) * 2022-11-21 2022-12-20 昆明理工大学 一种交流输电线路雷击故障识别方法及系统
CN115876475A (zh) * 2023-02-06 2023-03-31 山东大学 一种故障诊断方法、系统、设备及存储介质
CN115876475B (zh) * 2023-02-06 2023-06-09 山东大学 一种故障诊断方法、系统、设备及存储介质
CN116008731A (zh) * 2023-02-15 2023-04-25 重庆大学 配电网高阻故障辨识方法、装置和电子设备
CN116008731B (zh) * 2023-02-15 2023-08-25 重庆大学 配电网高阻故障辨识方法、装置和电子设备
CN116008756A (zh) * 2023-03-15 2023-04-25 国网福建省电力有限公司 电容式电压互感器绝缘故障诊断方法、系统、设备和介质
CN116087692A (zh) * 2023-04-12 2023-05-09 国网四川省电力公司电力科学研究院 一种配电网树线放电故障识别方法、系统、终端及介质
CN116827764A (zh) * 2023-08-23 2023-09-29 山西绿柳科技有限公司 一种基于神经网络的物联网故障检测控制方法及系统
CN116827764B (zh) * 2023-08-23 2023-11-03 山西绿柳科技有限公司 一种基于神经网络的物联网故障检测控制方法及系统
CN117368639A (zh) * 2023-10-18 2024-01-09 浙江德清迪生电力科技有限公司 基于人工智能的输电线路故障识别系统及方法
CN117250439B (zh) * 2023-11-08 2024-01-30 国网四川省电力公司电力科学研究院 一种面向多源接地故障的三层式研判分析系统
CN117250439A (zh) * 2023-11-08 2023-12-19 国网四川省电力公司电力科学研究院 一种面向多源接地故障的三层式研判分析系统
CN117216485A (zh) * 2023-11-09 2023-12-12 国网山东省电力公司电力科学研究院 一种基于客观赋权的输电录波鸟害故障判别方法及系统
CN117216485B (zh) * 2023-11-09 2024-01-30 国网山东省电力公司电力科学研究院 一种基于客观赋权的输电录波鸟害故障判别方法及系统
CN117313018A (zh) * 2023-11-29 2023-12-29 国网浙江省电力有限公司 一种电力输电线路异常状态识别方法及系统
CN117313018B (zh) * 2023-11-29 2024-01-30 国网浙江省电力有限公司 一种电力输电线路异常状态识别方法及系统
CN117310508A (zh) * 2023-11-30 2023-12-29 山东科技大学 一种快速准确测量锂电池电变量的方法
CN117310508B (zh) * 2023-11-30 2024-02-27 山东科技大学 一种快速准确测量锂电池电变量的方法
CN117851868A (zh) * 2024-01-15 2024-04-09 珠海安瑞通电子科技有限公司 基于深度学习的多模型推进式配电网故障识别方法及系统
CN117851868B (zh) * 2024-01-15 2024-06-11 珠海安瑞通电子科技有限公司 基于深度学习的多模型推进式配电网故障识别方法及系统

Also Published As

Publication number Publication date
CN113033837A (zh) 2021-06-25

Similar Documents

Publication Publication Date Title
WO2022183698A1 (zh) 基于输电线路暂态波形的人工智能故障辨识系统及方法
CN111680706B (zh) 一种基于编码和解码结构的双通道输出轮廓检测方法
CN110532900B (zh) 基于U-Net和LS-CNN的人脸表情识别方法
CN109949316A (zh) 一种基于rgb-t融合的电网设备图像弱监督实例分割方法
CN110309503A (zh) 一种基于深度学习bert--cnn的主观题评分模型及评分方法
CN107301376B (zh) 一种基于深度学习多层刺激的行人检测方法
CN107563433A (zh) 一种基于卷积神经网络的红外小目标检测方法
CN110930995B (zh) 一种应用于电力行业的语音识别模型
CN105023013B (zh) 基于局部标准差和Radon变换的目标检测方法
CN103177458A (zh) 一种基于频域分析的可见光遥感图像感兴趣区域检测方法
CN106344004A (zh) 心电信号特征点检测方法及装置
CN109284754A (zh) 基于计算机视觉的桥梁区域船舶几何轮廓自动辨识方法
CN109871757A (zh) 一种基于联合时频特征的雷达信号脉内调制类型识别方法
CN106778910A (zh) 基于本地训练的深度学习系统和方法
CN104504725A (zh) 一种快速自动语义图像分割模型方法
CN108182316A (zh) 一种基于人工智能的电磁仿真方法及其电磁大脑
Lin et al. A detection approach for bundled log ends using K-median clustering and improved YOLOv4-Tiny network
CN115050014A (zh) 一种基于图像文本学习的小样本番茄病害识别系统及方法
CN114758132B (zh) 一种基于卷积神经网络的果树病虫害识别方法及系统
CN112036419A (zh) 一种基于VGG-Attention模型的SAR图像部件解译方法
Zhang et al. Target detection of banana string and fruit stalk based on YOLOv3 deep learning network
Tang et al. Transound: Hyper-head attention transformer for birds sound recognition
He et al. Few-shot working condition recognition of a sucker-rod pumping system based on a 4-dimensional time-frequency signature and meta-learning convolutional shrinkage neural network
CN112329830A (zh) 一种基于卷积神经网络和迁移学习的无源定位轨迹数据识别方法及系统
Cai et al. An automatic trough line identification method based on improved UNet

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21928775

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21928775

Country of ref document: EP

Kind code of ref document: A1