WO2022001289A1 - Power distribution network partial discharge ultrasonic test method and system - Google Patents

Power distribution network partial discharge ultrasonic test method and system Download PDF

Info

Publication number
WO2022001289A1
WO2022001289A1 PCT/CN2021/087208 CN2021087208W WO2022001289A1 WO 2022001289 A1 WO2022001289 A1 WO 2022001289A1 CN 2021087208 W CN2021087208 W CN 2021087208W WO 2022001289 A1 WO2022001289 A1 WO 2022001289A1
Authority
WO
WIPO (PCT)
Prior art keywords
partial discharge
power distribution
neural network
equipment
feature
Prior art date
Application number
PCT/CN2021/087208
Other languages
French (fr)
Chinese (zh)
Inventor
张涛允
熊鹏
秦源汛
张广东
何红太
张玉刚
桂菲菲
白文远
王津
薛玲
张发刚
刘康
何卫锋
黄志勇
Original Assignee
北京国网富达科技发展有限责任公司
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 北京国网富达科技发展有限责任公司 filed Critical 北京国网富达科技发展有限责任公司
Priority to ZA2021/04616A priority Critical patent/ZA202104616B/en
Publication of WO2022001289A1 publication Critical patent/WO2022001289A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/0006—Industrial image inspection using a design-rule based approach
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10132—Ultrasound image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20052—Discrete cosine transform [DCT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • Y—GENERAL 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS 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
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Definitions

  • the invention relates to the technical field of distribution networks, in particular to a method and system for ultrasonic detection of partial discharges in distribution networks.
  • the purpose of the present invention is to provide a method for ultrasonic detection of partial discharge in distribution network based on deep learning, which solves the problems existing in the prior art and can efficiently and accurately detect the state of distribution network equipment.
  • the present invention provides the following scheme:
  • An ultrasonic detection method for partial discharge in a distribution network comprising:
  • the neural network model is trained according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment and the detection result of the historical power distribution equipment, and the trained neural network model is obtained;
  • the trained neural network model includes One layer of periodic neural network layer, one layer of convolutional neural network layer and multi-layer fully connected layer;
  • the third feature is input into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  • the neural network model is trained according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, and the trained neural network model is obtained, specifically:
  • adjusting the neural network model according to the output result and the historical power distribution equipment detection result to obtain a trained neural network model specifically:
  • the neural network model is adjusted by using the gradient back-propagation algorithm, so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
  • the detection result of the power distribution equipment includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the service life of the power distribution equipment.
  • the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested is converted into cepstrum data, specifically:
  • Windowing is performed on the digital signal to obtain a windowed digital signal
  • the cepstrum calculation is performed on the frequency signal to obtain Meyer's-eye cepstrum data.
  • a distribution network partial discharge ultrasonic detection system comprising:
  • the training module is used to train the neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, so as to obtain the trained neural network model;
  • the neural network model includes a periodic neural network layer, a convolutional neural network layer and a multi-layer fully connected layer;
  • the cepstrum data acquisition module is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into the cepstrum data;
  • a first feature acquisition module configured to input the cepstrum data into the periodic neural network layer for learning to obtain a first feature
  • the second feature acquisition module is used to input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature;
  • a third feature acquisition module configured to obtain a third feature after linearly splicing the first feature and the second feature
  • the detection module is used for inputting the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  • the training module includes:
  • an input unit configured to input the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result
  • An adjustment unit configured to adjust the neural network model according to the output result and the detection result of the historical power distribution equipment to obtain a trained neural network model.
  • the adjustment unit includes:
  • a judging sub-unit for judging whether the error between the output result and the historical power distribution equipment detection result is within the error range
  • the neural network model is adjusted by using the gradient back-propagation algorithm, so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
  • the detection result of the power distribution network equipment to be tested obtained by the detection module includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the service life of the power distribution equipment.
  • the cepstrum data acquisition module includes:
  • the digital signal conversion unit is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal;
  • a windowed digital signal acquisition unit configured to perform windowing processing on the digital signal to obtain a windowed digital signal
  • a signal sequence acquisition unit configured to perform Fourier transform on the windowed digital signal to obtain a signal sequence
  • a spectral signal acquisition unit configured to filter the signal sequence to obtain a filtered spectral signal
  • the cepstrum data acquisition unit is configured to perform cepstrum calculation on the frequency signal to obtain cepstrum data.
  • the present invention discloses the following technical effects:
  • the invention provides a method and system for ultrasonic detection of partial discharge in distribution network based on deep learning.
  • the method includes: training a neural network model; converting ultrasonic signals of partial discharge defects of distribution network equipment to be tested into cepstrum data ; Input the cepstrum data into the periodic neural network layer for learning to obtain the first feature; input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature; After the second feature is linearly spliced, the third feature is obtained; the third feature is input into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  • the detection method proposed by the present invention is more efficient and accurate.
  • FIG. 1 is a flowchart of a method for ultrasonic detection of partial discharge in a distribution network provided by an embodiment of the present invention
  • FIG. 2 is a process diagram of a cepstrum processing process provided by an embodiment of the present invention.
  • FIG. 3 is a structural diagram of a neural network model provided by an embodiment of the present invention.
  • FIG. 4 is a system block diagram of an ultrasonic detection system for partial discharge in a distribution network provided by an embodiment of the present invention.
  • the purpose of the present invention is to provide a method for ultrasonic detection of partial discharge in distribution network based on deep learning, which solves the problems existing in the prior art and can efficiently and accurately detect the state of distribution network equipment.
  • FIG. 1 is a flowchart of a method for ultrasonic detection of partial discharge in a distribution network provided by an embodiment of the present invention. As shown in Figure 1, in this embodiment, the method includes the following steps:
  • Step 101 Train a neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, and obtain a trained neural network model;
  • the network model includes a periodic neural network layer, a convolutional neural network layer and a multi-layer fully connected layer.
  • the training process of the neural network model includes the following steps:
  • Step 1011 Input the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain the output result.
  • the output results include the defect type of the partial discharge of the power distribution equipment, the severity of the power distribution equipment failure and the service life of the power distribution equipment.
  • Step 1012 Adjust the neural network model according to the output results and the historical power distribution equipment detection results. Determine whether the error between the output result and the historical power distribution equipment detection result is within the error range. If so, determine that the neural network model is a trained neural network model. If not, use the gradient back-propagation algorithm to adjust the neural network model so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
  • Step 102 Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into cepstrum data.
  • FIG. 2 is a process diagram of a cepstrum processing process provided by an embodiment of the present invention.
  • converting the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into cepstrum data includes the following steps:
  • Step 1021 Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal.
  • Step 1022 Windowing the digital signal to obtain a windowed digital signal.
  • a[n] is the digital signal after windowing
  • s[n] is the digital signal before windowing
  • w[n] is the formula of Hanning window
  • a is the Hanning window at 0.46164
  • ⁇ is 0.5
  • L is the width of the window
  • n is the number of ultrasonic signal frames.
  • Step 1023 Perform Fourier transform on the windowed digital signal to obtain a signal sequence.
  • the Fourier transform formula is as follows:
  • a[k] is the signal sequence after Fourier transform
  • k is the subscript of the frame number of the signal sequence
  • j is the imaginary part
  • N is the total number of frames of the signal sequence.
  • Step 1024 Filter the signal sequence to obtain a filtered spectrum signal.
  • the filtering formula is as follows:
  • X t [m] is the filtered spectrum signal
  • W is the filter group
  • m is a frequency band filter is a frequency domain index
  • k is the spectrum index signal frame
  • a t represents the signal intensity normalized vector.
  • Step 1025 Perform cepstral calculation on the frequency signal to obtain Meier's cepstral data.
  • the specific calculation formula is:
  • x t [n] is the cepstral data of Merlin
  • M is the total number of filter channels.
  • Step 103 Input the cepstrum data into the periodic neural network layer for learning to obtain a first feature.
  • Step 104 Input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature.
  • Step 105 Linearly splicing the first feature and the second feature to obtain a third feature.
  • Step 106 Input the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  • FIG. 3 is a structural diagram of a neural network model provided by an embodiment of the present invention, as shown in FIG. 3 :
  • s t represents the state, and s t depends on the current input x t and the s t-1 of the previous hidden layer.
  • the weight matrix W is the previous value of the hidden layer as the weight of this input.
  • U represents the input normalized vector parameter, and V represents the output normalized vector parameter.
  • the output vector o t represents the confidence of equipment defect classification for each fault, and artificial intelligence uses different types of confidence to estimate equipment defect types.
  • the weight matrix W can be continuously corrected by the gradient descent method in the multiple training process, and the trained W is the model we need.
  • the defect types of partial discharge of power distribution equipment, the severity of power distribution equipment faults and the service life of power distribution equipment are obtained.
  • FIG. 4 is a system block diagram of an ultrasonic detection system for partial discharge in a distribution network provided by an embodiment of the present invention. As shown in Figure 4, the system includes:
  • the training module 201 is used to train a neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, so as to obtain a trained neural network model;
  • the training A good neural network model includes a periodic neural network layer, a convolutional neural network layer, and a multi-layer fully connected layer.
  • the training module 201 specifically includes:
  • the input unit 2011 is used for inputting the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result.
  • the adjustment unit 2012 is configured to adjust the neural network model according to the output result and the detection result of the historical power distribution equipment to obtain a trained neural network model.
  • the adjusting unit 2012 includes a judging subunit, which is used to judge whether the error between the output result and the historical power distribution equipment detection result is within the error range, and if so, determine that the neural network model is a trained neural network model. If not, use the gradient back-propagation algorithm to adjust the neural network model so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
  • the cepstrum data acquisition module 202 is configured to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into the cepstrum data.
  • the cepstrum data acquisition module 202 specifically includes:
  • the digital signal conversion unit 2021 is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal.
  • the windowed digital signal acquisition unit 2022 is configured to perform windowing processing on the digital signal to obtain a windowed digital signal.
  • the signal sequence acquisition unit 2023 is configured to perform Fourier transform on the windowed digital signal to obtain a signal sequence.
  • the spectral signal obtaining unit 2024 is configured to filter the signal sequence to obtain the filtered spectral signal.
  • the cepstrum data acquisition unit 2025 is configured to perform cepstrum calculation on the frequency signal to obtain cepstrum data.
  • the first feature acquisition module 203 is configured to input the cepstrum data into the periodic neural network layer for learning to obtain a first feature.
  • the second feature acquisition module 204 is configured to input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature.
  • the third feature acquisition module 205 is configured to obtain a third feature after linearly splicing the first feature and the second feature.
  • the detection module 206 is configured to input the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  • the detection result includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the service life of the power distribution equipment.
  • the present invention discloses the following technical effects:
  • the present invention proposes a method and system for ultrasonic detection of partial discharge in power distribution network based on deep learning. According to a large number of ultrasonic signals of partial discharge defects of historical power distribution equipment, images of partial discharge defects of historical power distribution equipment and historical power distribution equipment detection As a result, the neural network model is trained. When detecting distribution network equipment, only the ultrasonic signals and images of partial discharge defects of the power distribution equipment to be tested are input into the trained neural network model, and the detection results of the power distribution equipment to be tested can be obtained. Compared with the existing manual detection, it is more efficient and accurate.
  • Operation and maintenance personnel can discover equipment defects and hidden dangers endangering line safety in advance, grasp the operating conditions of line equipment in time, and then take targeted control measures to effectively reduce the number of line power outages and maintenance, ensure the safe and stable operation of distribution lines, and effectively promote distribution lines.
  • the trained neural network model can be pruned and transplanted to the front-end ultrasonic detector to realize real-time partial discharge type diagnosis and classification, realize the intelligent and refined inspection equipment, and greatly improve the inspection efficiency.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Testing Relating To Insulation (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)

Abstract

The present invention relates to a deep learning-based power distribution network partial discharge ultrasonic test method and system. The method comprises: training a neural network model; converting an ultrasonic signal of a partial discharge defect of a power distribution network device to be tested into Mel Frequency Cepstral data; inputting the Mel Frequency Cepstral data into a periodic neural network layer for learning to obtain a first feature; inputting an image of the partial discharge defect of the power distribution network device to be tested into a convolutional neural network layer for learning to obtain a second feature; linearly stitching the first feature and the second feature to obtain a third feature; and inputting the third feature into a multi-layer full connection layer to obtain a test result of the power distribution network device to be tested. Compared with the existing manual tests, the test method and system provided in the present invention are more efficient and accurate.

Description

一种配电网局部放电超声波检测方法及系统Method and system for ultrasonic detection of partial discharge in distribution network
本申请要求于2020年06月28日提交中国专利局、申请号为202010596194.1、发明名称为“一种配电网局部放电超声波检测方法及系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。This application claims the priority of the Chinese patent application filed on June 28, 2020 with the application number 202010596194.1 and the invention titled "A method and system for ultrasonic detection of partial discharge in a distribution network", the entire contents of which are by reference Incorporated in this application.
技术领域technical field
本发明涉及配电网技术领域，特别是涉及一种配电网局部放电超声波检测方法及系统。The invention relates to the technical field of distribution networks, in particular to a method and system for ultrasonic detection of partial discharges in distribution networks.
背景技术Background technique
配电线路巡视工作是配电专业日常运维管理的重要工作。我国配电线路以架空方式为主，在线路运行过程中，高压电气设备长期存在局部放电，会加速设备老化，最终导致故障发生。架空线路设备发生局部放电时仅凭巡视人员肉眼与耳朵很难发现，特别是一些轻微的局部放电。目前我国电力企业对生产运维精益化要求不断提高，新设备、新材料的持续增长以及配网设备整体规模数量的急剧增加，导致现场检修、日常运行维护工作剧增，生产结构性缺员与供电可靠性要求的矛盾日益突出。The inspection of power distribution lines is an important task in the daily operation and maintenance management of power distribution professionals. my country's distribution lines are mainly overhead. During the operation of the line, there is a long-term partial discharge in the high-voltage electrical equipment, which will accelerate the aging of the equipment and eventually lead to failure. When partial discharge occurs in overhead line equipment, it is difficult to detect only with the naked eyes and ears of inspectors, especially some slight partial discharges. At present, my country's electric power enterprises have continuously increased the requirements for lean production, operation and maintenance, the continuous growth of new equipment and new materials, and the sharp increase in the overall scale and number of distribution network equipment, resulting in a sharp increase in on-site maintenance, daily operation and maintenance work, and structural shortages in production. The contradiction of power supply reliability requirements has become increasingly prominent.
传统的通过外观检查、手工记录开展配电线路日常巡视检查的工作方式，无法对设备状态及潜伏性故障有效掌握，特别是当前我国对输配电设备施行定期检修等方式，针对性不强。传统检测无法掌握设备的潜伏性故障，对设备状态难以真实有效地进行评价，以致检修策略缺乏针对性，可能导致设备“失修”、“过修”的问题频频出现；同时传统检修增加了停电次数。传统的计划检修和例行试验需要对线路进行停电，客观上降低了供电可靠性指标，也可能因检修周期过长的影响无法及时掌控设备状态。The traditional way of conducting daily inspection and inspection of distribution lines through visual inspection and manual recording cannot effectively grasp the equipment status and latent faults. Traditional detection cannot grasp the latent faults of equipment, and it is difficult to evaluate equipment status truly and effectively, resulting in lack of targeted maintenance strategies, which may lead to frequent occurrence of equipment "disrepair" and "over-repair" problems; at the same time, traditional maintenance increases the number of power outages . Traditional planned maintenance and routine tests require power outages on the line, which objectively reduces the reliability of power supply, and may not be able to control the equipment status in time due to the long maintenance cycle.
发明内容SUMMARY OF THE INVENTION
本发明的目的是提供一种基于深度学习的配电网局部放电超声波检测方法，解决了现有技术存在的问题，能够高效、准确的检测配网设备的状态。The purpose of the present invention is to provide a method for ultrasonic detection of partial discharge in distribution network based on deep learning, which solves the problems existing in the prior art and can efficiently and accurately detect the state of distribution network equipment.
为实现上述目的，本发明提供了如下方案：For achieving the above object, the present invention provides the following scheme:
一种配电网局部放电超声波检测方法，包括：An ultrasonic detection method for partial discharge in a distribution network, comprising:
根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型；所述训练好的神经网络模型包括一层周期神经网络层、一层卷积神经网络层和多层全连接层；The neural network model is trained according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment and the detection result of the historical power distribution equipment, and the trained neural network model is obtained; the trained neural network model includes One layer of periodic neural network layer, one layer of convolutional neural network layer and multi-layer fully connected layer;
将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据；Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into cepstrum data;
将所述梅氏倒频谱数据输入所述周期神经网络层进行学习得到第一特征；Inputting the cepstrum data into the periodic neural network layer for learning to obtain the first feature;
将待测配电网设备的局部放电缺陷的图像输入所述卷积神经网络层进行学习得到第二特征；Inputting the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature;
将所述第一特征和所述第二特征进行线性拼接后得到第三特征；After linearly splicing the first feature and the second feature, a third feature is obtained;
将所述第三特征输入所述多层全连接层，得到待测配电网设备的检测结果。The third feature is input into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
可选的，所述根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型，具体为：Optionally, the neural network model is trained according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, and the trained neural network model is obtained, specifically:
将所述历史配电设备局部放电缺陷的超声波信号和所述历史配电设备局部放电缺陷的图像输入至所述神经网络模型，得到输出结果；Inputting the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result;
根据所述输出结果以及所述历史配电设备检测结果调整所述神经网络模型，得到训练好的神经网络模型。Adjust the neural network model according to the output result and the historical power distribution equipment detection result to obtain a trained neural network model.
可选的，所述根据所述输出结果以及所述历史配电设备检测结果调整所述神经网络模型，得到训练好的神经网络模型，具体为：Optionally, adjusting the neural network model according to the output result and the historical power distribution equipment detection result to obtain a trained neural network model, specifically:
判断所述输出结果与所述历史配电设备检测结果的误差是否在误差范围内；Judging whether the error between the output result and the historical power distribution equipment detection result is within the error range;
若是，则确定所述神经网络模型为训练好的神经网络模型；If so, then determine that the neural network model is a trained neural network model;
若否，则利用梯度反向传播算法调整所述神经网络模型，使所述输出结果与所述历史配电设备检测结果的误差在误差范围内。If not, the neural network model is adjusted by using the gradient back-propagation algorithm, so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
可选的，所述配电设备检测结果包括配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。Optionally, the detection result of the power distribution equipment includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the service life of the power distribution equipment.
可选的，所述将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏 倒频谱数据，具体为：Optionally, the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested is converted into cepstrum data, specifically:
将待测配电网设备的局部放电缺陷的超声波信号转换成数字信号；Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal;
对所述数字信号进行加窗处理获得加窗后的数字信号；Windowing is performed on the digital signal to obtain a windowed digital signal;
将所述加窗后的数字信号进行傅里叶变换，得到信号序列；Fourier transform is performed on the digital signal after the windowing to obtain a signal sequence;
对所述信号序列进行滤波，得到滤波后的频谱信号；Filtering the signal sequence to obtain a filtered spectrum signal;
对所述频率信号进行倒频谱计算得到梅氏倒频谱数据。The cepstrum calculation is performed on the frequency signal to obtain Meyer's-eye cepstrum data.
一种配电网局部放电超声波检测系统，包括：A distribution network partial discharge ultrasonic detection system, comprising:
训练模块，用于根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型；所述训练好的神经网络模型包括一层周期神经网络层、一层卷积神经网络层和多层全连接层；The training module is used to train the neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, so as to obtain the trained neural network model; The neural network model includes a periodic neural network layer, a convolutional neural network layer and a multi-layer fully connected layer;
梅氏倒频谱数据获取模块，用于将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据；The cepstrum data acquisition module is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into the cepstrum data;
第一特征获取模块，用于将所述梅氏倒频谱数据输入所述周期神经网络层进行学习得到第一特征；a first feature acquisition module, configured to input the cepstrum data into the periodic neural network layer for learning to obtain a first feature;
第二特征获取模块，用于将待测配电网设备的局部放电缺陷的图像输入所述卷积神经网络层进行学习得到第二特征；The second feature acquisition module is used to input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature;
第三特征获取模块，用于将所述第一特征和所述第二特征进行线性拼接后得到第三特征；A third feature acquisition module, configured to obtain a third feature after linearly splicing the first feature and the second feature;
检测模块，用于将所述第三特征输入所述多层全连接层，得到待测配电网设备的检测结果。The detection module is used for inputting the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
可选的，所述训练模块包括：Optionally, the training module includes:
输入单元，用于将所述历史配电设备局部放电缺陷的超声波信号和所述历史配电设备局部放电缺陷的图像输入至所述神经网络模型，得到输出结果；an input unit, configured to input the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result;
调整单元，用于根据所述输出结果以及所述历史配电设备检测结果调整所述神经网络模型，得到训练好的神经网络模型。An adjustment unit, configured to adjust the neural network model according to the output result and the detection result of the historical power distribution equipment to obtain a trained neural network model.
可选的，所述调整单元包括：Optionally, the adjustment unit includes:
判断子单元，用于判断所述输出结果与所述历史配电设备检测结果的误差是否在误差范围内；a judging sub-unit for judging whether the error between the output result and the historical power distribution equipment detection result is within the error range;
若是，则确定所述神经网络模型为训练好的神经网络模型；If so, then determine that the neural network model is a trained neural network model;
若否，则利用梯度反向传播算法调整所述神经网络模型，使所述输出结果与所述历史配电设备检测结果的误差在误差范围内。If not, the neural network model is adjusted by using the gradient back-propagation algorithm, so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
可选的，所述检测模块得到的所述待测配电网设备的检测结果包括配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。Optionally, the detection result of the power distribution network equipment to be tested obtained by the detection module includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the service life of the power distribution equipment.
可选的，所述梅氏倒频谱数据获取模块包括：Optionally, the cepstrum data acquisition module includes:
数字信号转换单元，用于将待测配电网设备的局部放电缺陷的超声波信号转换成数字信号；The digital signal conversion unit is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal;
加窗数字信号获取单元，用于对所述数字信号进行加窗处理获得加窗后的数字信号；a windowed digital signal acquisition unit, configured to perform windowing processing on the digital signal to obtain a windowed digital signal;
信号序列获取单元，用于将所述加窗后的数字信号进行傅里叶变换，得到信号序列；a signal sequence acquisition unit, configured to perform Fourier transform on the windowed digital signal to obtain a signal sequence;
频谱信号获取单元，用于对所述信号序列进行滤波，得到滤波后的频谱信号；a spectral signal acquisition unit, configured to filter the signal sequence to obtain a filtered spectral signal;
梅氏倒频谱数据获取单元，用于对所述频率信号进行倒频谱计算得到梅氏倒频谱数据。The cepstrum data acquisition unit is configured to perform cepstrum calculation on the frequency signal to obtain cepstrum data.
根据本发明提供的具体实施例，本发明公开了以下技术效果：According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
本发明提出了一种基于深度学习的配电网局部放电超声波检测方法及系统，方法包括：训练神经网络模型；将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据；将梅氏倒频谱数据输入周期神经网络层进行学习得到第一特征；将待测配电网设备的局部放电缺陷的图像输入卷积神经网络层进行学习得到第二特征；将第一特征和第二特征进行线性拼接后得到第三特征；将第三特征输入多层全连接层，得到待测配电网设备的检测结果。本发明提出的检测方法相对于现有的人工检测更高效、更准确。The invention provides a method and system for ultrasonic detection of partial discharge in distribution network based on deep learning. The method includes: training a neural network model; converting ultrasonic signals of partial discharge defects of distribution network equipment to be tested into cepstrum data ; Input the cepstrum data into the periodic neural network layer for learning to obtain the first feature; input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature; After the second feature is linearly spliced, the third feature is obtained; the third feature is input into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested. Compared with the existing manual detection, the detection method proposed by the present invention is more efficient and accurate.
附图说明Description of drawings
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings required in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the present invention. In the embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative labor.
图1为本发明实施例提供的配电网局部放电超声波检测方法的流程图；1 is a flowchart of a method for ultrasonic detection of partial discharge in a distribution network provided by an embodiment of the present invention;
图2为本发明实施例提供的梅氏倒频谱处理过程图；FIG. 2 is a process diagram of a cepstrum processing process provided by an embodiment of the present invention;
图3为本发明实施例提供的神经网络模型结构图；3 is a structural diagram of a neural network model provided by an embodiment of the present invention;
图4为本发明实施例提供的配电网局部放电超声波检测系统的系统框图。FIG. 4 is a system block diagram of an ultrasonic detection system for partial discharge in a distribution network provided by an embodiment of the present invention.
具体实施方式detailed description
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
本发明的目的是提供一种基于深度学习的配电网局部放电超声波检测方法，解决了现有技术存在的问题，能够高效、准确的检测配网设备的状态。The purpose of the present invention is to provide a method for ultrasonic detection of partial discharge in distribution network based on deep learning, which solves the problems existing in the prior art and can efficiently and accurately detect the state of distribution network equipment.
为使本发明的上述目的、特征和优点能够更加明显易懂，下面结合附图和具体实施方式对本发明作进一步详细的说明。In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.
实施例1Example 1
图1为本发明实施例提供的配电网局部放电超声波检测方法的流程图。如图1所示，在本实施例中，方法包括以下步骤：FIG. 1 is a flowchart of a method for ultrasonic detection of partial discharge in a distribution network provided by an embodiment of the present invention. As shown in Figure 1, in this embodiment, the method includes the following steps:
步骤101：根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型；所述训练好的神经网络模型包括一层周期神经网络层、一层卷积神经网络层和多层全连接层。Step 101: Train a neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, and obtain a trained neural network model; The network model includes a periodic neural network layer, a convolutional neural network layer and a multi-layer fully connected layer.
在本实施例中，神经网络模型的训练过程包括以下步骤：In this embodiment, the training process of the neural network model includes the following steps:
步骤1011：将历史配电设备局部放电缺陷的超声波信号和历史配电设备局部放电缺陷的图像输入至神经网络模型，得到输出结果。输出结果包括配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。Step 1011: Input the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain the output result. The output results include the defect type of the partial discharge of the power distribution equipment, the severity of the power distribution equipment failure and the service life of the power distribution equipment.
步骤1012：根据输出结果以及历史配电设备检测结果调整神经网络模型。判断输出结果与历史配电设备检测结果的误差是否在误差范围内。若是，则确定此神经网络模型为训练好的神经网络模型。若否，则利用梯度反向传播算法调整神经网络模型，使输出结果与历史配电设备检测结果的误差在误差范围内。Step 1012: Adjust the neural network model according to the output results and the historical power distribution equipment detection results. Determine whether the error between the output result and the historical power distribution equipment detection result is within the error range. If so, determine that the neural network model is a trained neural network model. If not, use the gradient back-propagation algorithm to adjust the neural network model so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
步骤102：将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据。Step 102: Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into cepstrum data.
图2为本发明实施例提供的梅氏倒频谱处理过程图。在本实施例中，将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据，包括以下步骤：FIG. 2 is a process diagram of a cepstrum processing process provided by an embodiment of the present invention. In this embodiment, converting the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into cepstrum data includes the following steps:
步骤1021：将待测配电网设备的局部放电缺陷的超声波信号转换成数字信号。Step 1021: Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal.
步骤1022：对数字信号进行加窗处理获得加窗后的数字信号。Step 1022: Windowing the digital signal to obtain a windowed digital signal.
在本实施例中，利用公式In this embodiment, using the formula
a[n]＝w[n]*s[n]a[n]=w[n]*s[n]
进行加窗计算。其中a[n]为加窗后的数字信号，s[n]为加窗前的数字信号，w[n]为汉宁窗的公式，
Figure PCTCN2021087208-appb-000001
a为0.46164时的汉宁窗，α取0.5，L为窗的宽度，n为超声波信号帧数。 Perform windowing calculations. where a[n] is the digital signal after windowing, s[n] is the digital signal before windowing, w[n] is the formula of Hanning window,
Figure PCTCN2021087208-appb-000001
a is the Hanning window at 0.46164, α is 0.5, L is the width of the window, and n is the number of ultrasonic signal frames.
步骤1023：将加窗后的数字信号进行傅里叶变换，得到信号序列。傅里叶变换公式如下：Step 1023: Perform Fourier transform on the windowed digital signal to obtain a signal sequence. The Fourier transform formula is as follows:
Figure PCTCN2021087208-appb-000002
Figure PCTCN2021087208-appb-000002
其中a[k]为经过傅里叶变换后的信号序列，k为信号序列的帧数下标，j为虚部，N为信号序列的总帧数。Where a[k] is the signal sequence after Fourier transform, k is the subscript of the frame number of the signal sequence, j is the imaginary part, and N is the total number of frames of the signal sequence.
步骤1024：对信号序列进行滤波，得到滤波后的频谱信号。滤波公式如下：Step 1024: Filter the signal sequence to obtain a filtered spectrum signal. The filtering formula is as follows:
Figure PCTCN2021087208-appb-000003
Figure PCTCN2021087208-appb-000003
其中，X t[m]为滤波后的频谱信号，W为滤波器组，m为频域滤波器的频带下标，k为频谱信号的帧下标，A t表示信号强度归一化向量。 Wherein, X t [m] is the filtered spectrum signal, W is the filter group, m is a frequency band filter is a frequency domain index, k is the spectrum index signal frame, A t represents the signal intensity normalized vector.
步骤1025：对频率信号进行倒频谱计算得到梅氏倒频谱数据。具体计算公式为：Step 1025 : Perform cepstral calculation on the frequency signal to obtain Meier's cepstral data. The specific calculation formula is:
Figure PCTCN2021087208-appb-000004
Figure PCTCN2021087208-appb-000004
其中，x t[n]为梅氏倒频谱数据，M为滤波器通道总数。 Among them, x t [n] is the cepstral data of Merlin, and M is the total number of filter channels.
步骤103：将所述梅氏倒频谱数据输入所述周期神经网络层进行学习得到第一特征。Step 103: Input the cepstrum data into the periodic neural network layer for learning to obtain a first feature.
步骤104：将待测配电网设备的局部放电缺陷的图像输入所述卷积神经网络层进行学习得到第二特征。Step 104: Input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature.
步骤105：将所述第一特征和所述第二特征进行线性拼接后得到第三特征。Step 105: Linearly splicing the first feature and the second feature to obtain a third feature.
步骤106：将所述第三特征输入所述多层全连接层，得到待测配电网设备的检测结果。Step 106: Input the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
图3为本发明实施例提供的神经网络模型结构图，如图3所示：FIG. 3 is a structural diagram of a neural network model provided by an embodiment of the present invention, as shown in FIG. 3 :
图中s t表示状态，s t取决于当前输入x t以及上一次隐藏层的s t-1，权重矩阵W就是隐藏层上一次的值作为这一次的输入的权重。U表示输入的归一化向量参数，V表示输出的归一化向量参数。计算过程可以表示为： In the figure, s t represents the state, and s t depends on the current input x t and the s t-1 of the previous hidden layer. The weight matrix W is the previous value of the hidden layer as the weight of this input. U represents the input normalized vector parameter, and V represents the output normalized vector parameter. The calculation process can be expressed as:
s t＝f(U*X t+W*s t-1) s t =f(U*X t +W*s t-1 )
o t＝g(V*s t) o t =g(V*s t )
其中输出向量o t表示的是设备缺陷为各个故障分类的置信度，人工智能使用不同类型的置信度来估计设备缺陷类型。通过深度学习的梯度反向传播算法，可以在多次训练过程中，通过梯度下降的方法不断修正权重矩阵W，训练好的W即为我们需要的模型。将待测配电设备的局部放电缺陷的超声波信号和局部放电缺陷的图像输入W中，可得出新的o t，即各个可能缺陷类型的置信程度，取最高值即可得出估计的缺陷的分类值，最终得出配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。 The output vector o t represents the confidence of equipment defect classification for each fault, and artificial intelligence uses different types of confidence to estimate equipment defect types. Through the gradient back-propagation algorithm of deep learning, the weight matrix W can be continuously corrected by the gradient descent method in the multiple training process, and the trained W is the model we need. Input the ultrasonic signal of the partial discharge defect of the power distribution equipment to be tested and the image of the partial discharge defect into W, a new o t can be obtained, that is, the confidence level of each possible defect type, and the estimated defect can be obtained by taking the highest value Finally, the defect types of partial discharge of power distribution equipment, the severity of power distribution equipment faults and the service life of power distribution equipment are obtained.
实施例2Example 2
为了能够高效、准确的检测配网设备的状态，本实施例还提供了一种配电网局部放电超声波检测系统。图4为本发明实施例提供的配电网局部放电超声波检测系统的系统框图。如图4所示，系统包括：In order to efficiently and accurately detect the state of the distribution network equipment, this embodiment also provides an ultrasonic detection system for partial discharge in the distribution network. FIG. 4 is a system block diagram of an ultrasonic detection system for partial discharge in a distribution network provided by an embodiment of the present invention. As shown in Figure 4, the system includes:
训练模块201，用于根据历史配电设备局部放电缺陷的超声波信号、历史 配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型；所述训练好的神经网络模型包括一层周期神经网络层、一层卷积神经网络层和多层全连接层。The training module 201 is used to train a neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment, so as to obtain a trained neural network model; the training A good neural network model includes a periodic neural network layer, a convolutional neural network layer, and a multi-layer fully connected layer.
在本实施例中，训练模块201具体包括：In this embodiment, the training module 201 specifically includes:
输入单元2011，用于将历史配电设备局部放电缺陷的超声波信号和历史配电设备局部放电缺陷的图像输入至神经网络模型，得到输出结果。The input unit 2011 is used for inputting the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result.
调整单元2012，用于根据输出结果以及历史配电设备检测结果调整神经网络模型，得到训练好的神经网络模型。具体的，调整单元2012包括判断子单元，判断子单元用于判断输出结果与历史配电设备检测结果的误差是否在误差范围内，若是，则确定此神经网络模型为训练好的神经网络模型。若否，则利用梯度反向传播算法调整神经网络模型，使输出结果与历史配电设备检测结果的误差在误差范围内。The adjustment unit 2012 is configured to adjust the neural network model according to the output result and the detection result of the historical power distribution equipment to obtain a trained neural network model. Specifically, the adjusting unit 2012 includes a judging subunit, which is used to judge whether the error between the output result and the historical power distribution equipment detection result is within the error range, and if so, determine that the neural network model is a trained neural network model. If not, use the gradient back-propagation algorithm to adjust the neural network model so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
梅氏倒频谱数据获取模块202，用于将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据。The cepstrum data acquisition module 202 is configured to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into the cepstrum data.
在本实施例中，梅氏倒频谱数据获取模块202具体包括：In this embodiment, the cepstrum data acquisition module 202 specifically includes:
数字信号转换单元2021，用于将待测配电网设备的局部放电缺陷的超声波信号转换成数字信号。The digital signal conversion unit 2021 is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal.
加窗数字信号获取单元2022，用于对数字信号进行加窗处理获得加窗后的数字信号。The windowed digital signal acquisition unit 2022 is configured to perform windowing processing on the digital signal to obtain a windowed digital signal.
信号序列获取单元2023，用于将加窗后的数字信号进行傅里叶变换，得到信号序列。The signal sequence acquisition unit 2023 is configured to perform Fourier transform on the windowed digital signal to obtain a signal sequence.
频谱信号获取单元2024，用于对信号序列进行滤波，得到滤波后的频谱信号。The spectral signal obtaining unit 2024 is configured to filter the signal sequence to obtain the filtered spectral signal.
梅氏倒频谱数据获取单元2025，用于对频率信号进行倒频谱计算得到梅氏倒频谱数据。The cepstrum data acquisition unit 2025 is configured to perform cepstrum calculation on the frequency signal to obtain cepstrum data.
第一特征获取模块203，用于将所述梅氏倒频谱数据输入所述周期神经网络层进行学习得到第一特征。The first feature acquisition module 203 is configured to input the cepstrum data into the periodic neural network layer for learning to obtain a first feature.
第二特征获取模块204，用于将待测配电网设备的局部放电缺陷的图像输入所述卷积神经网络层进行学习得到第二特征。The second feature acquisition module 204 is configured to input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature.
第三特征获取模块205，用于将所述第一特征和所述第二特征进行线性拼接后得到第三特征。The third feature acquisition module 205 is configured to obtain a third feature after linearly splicing the first feature and the second feature.
检测模块206，用于将所述第三特征输入所述多层全连接层，得到待测配电网设备的检测结果。在本实施例中，检测结果包括配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。The detection module 206 is configured to input the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested. In this embodiment, the detection result includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the service life of the power distribution equipment.
根据本发明提供的具体实施例，本发明公开了以下技术效果：According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
本发明提出了一种基于深度学习的配电网局部放电超声波检测方法及系统，根据大量的历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果，来训练神经网络模型。检测配电网设备时，只需要将待测配电设备局部放电缺陷的超声波信号和图像输入训练好的神经网络模型中，即可得出待测配电设备的检测结果。相对于现有的人工检测更高效、更准确。运维人员可提前发现设备缺陷和危及线路安全的隐患，及时掌握线路设备运行工况，进而采取有针对性的治理措施，有效减少线路停电检修次数，确保配电线路安全稳定运行，有力促进配网供电可靠性指标的快速提升。The present invention proposes a method and system for ultrasonic detection of partial discharge in power distribution network based on deep learning. According to a large number of ultrasonic signals of partial discharge defects of historical power distribution equipment, images of partial discharge defects of historical power distribution equipment and historical power distribution equipment detection As a result, the neural network model is trained. When detecting distribution network equipment, only the ultrasonic signals and images of partial discharge defects of the power distribution equipment to be tested are input into the trained neural network model, and the detection results of the power distribution equipment to be tested can be obtained. Compared with the existing manual detection, it is more efficient and accurate. Operation and maintenance personnel can discover equipment defects and hidden dangers endangering line safety in advance, grasp the operating conditions of line equipment in time, and then take targeted control measures to effectively reduce the number of line power outages and maintenance, ensure the safe and stable operation of distribution lines, and effectively promote distribution lines. The rapid improvement of network power supply reliability indicators.
而且可以将训练好的神经网络模型经过修剪移植到前端的超声波检测仪中实现实时的局部放电类型诊断分类，实现巡检设备的智能化精细化，进而大幅提高巡检效率。In addition, the trained neural network model can be pruned and transplanted to the front-end ultrasonic detector to realize real-time partial discharge type diagnosis and classification, realize the intelligent and refined inspection equipment, and greatly improve the inspection efficiency.
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other.
本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处。综上所述，本说明书内容不应理解为对本发明的限制。The principles and implementations of the present invention are described herein using specific examples, and the descriptions of the above embodiments are only used to help understand the core idea of the present invention; There will be changes in the specific implementation manner and application scope. In conclusion, the contents of this specification should not be construed as limiting the present invention.

Claims (10)

  1. 一种配电网局部放电超声波检测方法，其特征在于，包括：A method for ultrasonic detection of partial discharge in a distribution network, comprising:
    根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型；所述训练好的神经网络模型包括一层周期神经网络层、一层卷积神经网络层和多层全连接层；The neural network model is trained according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment and the detection result of the historical power distribution equipment, and the trained neural network model is obtained; the trained neural network model includes One layer of periodic neural network layer, one layer of convolutional neural network layer and multi-layer fully connected layer;
    将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据；Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into cepstrum data;
    将所述梅氏倒频谱数据输入所述周期神经网络层进行学习得到第一特征；Inputting the cepstrum data into the periodic neural network layer for learning to obtain the first feature;
    将待测配电网设备的局部放电缺陷的图像输入所述卷积神经网络层进行学习得到第二特征；Inputting the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature;
    将所述第一特征和所述第二特征进行线性拼接后得到第三特征；After linearly splicing the first feature and the second feature, a third feature is obtained;
    将所述第三特征输入所述多层全连接层，得到待测配电网设备的检测结果。The third feature is input into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  2. 根据权利要求1所述的配电网局部放电超声波检测方法，其特征在于，所述根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型，具体为：The method for ultrasonic detection of partial discharge in a distribution network according to claim 1, wherein the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment, and the detection result of the historical power distribution equipment Train the neural network model to get the trained neural network model, specifically:
    将所述历史配电设备局部放电缺陷的超声波信号和所述历史配电设备局部放电缺陷的图像输入至所述神经网络模型，得到输出结果；Inputting the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result;
    根据所述输出结果以及所述历史配电设备检测结果调整所述神经网络模型，得到训练好的神经网络模型。Adjust the neural network model according to the output result and the historical power distribution equipment detection result to obtain a trained neural network model.
  3. 根据权利要求2所述的配电网局部放电超声波检测方法，其特征在于，所述根据所述输出结果以及所述历史配电设备检测结果调整所述神经网络模型，得到训练好的神经网络模型，具体为：The method for ultrasonic detection of partial discharge in a power distribution network according to claim 2, wherein the neural network model is adjusted according to the output result and the detection result of the historical power distribution equipment to obtain a trained neural network model ,Specifically:
    判断所述输出结果与所述历史配电设备检测结果的误差是否在误差范围内；Judging whether the error between the output result and the historical power distribution equipment detection result is within the error range;
    若是，则确定所述神经网络模型为训练好的神经网络模型；If so, then determine that the neural network model is a trained neural network model;
    若否，则利用梯度反向传播算法调整所述神经网络模型，使所述输出结果与所述历史配电设备检测结果的误差在误差范围内。If not, the neural network model is adjusted by using the gradient back-propagation algorithm, so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
  4. 根据权利要求1所述的配电网局部放电超声波检测方法，其特征在于，所述配电设备检测结果包括配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。The method for ultrasonic detection of partial discharge in a distribution network according to claim 1, wherein the detection result of the power distribution equipment includes the defect type of the partial discharge of the power distribution equipment, the severity of the fault of the power distribution equipment, and the use of the power distribution equipment. life.
  5. 根据权利要求1所述的配电网局部放电超声波检测方法，其特征在于，所述将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据，具体为：The method for ultrasonic detection of partial discharge in a distribution network according to claim 1, wherein the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested is converted into cepstrum data, specifically:
    将待测配电网设备的局部放电缺陷的超声波信号转换成数字信号；Convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal;
    对所述数字信号进行加窗处理获得加窗后的数字信号；Windowing is performed on the digital signal to obtain a windowed digital signal;
    将所述加窗后的数字信号进行傅里叶变换，得到信号序列；Fourier transform is performed on the digital signal after the windowing to obtain a signal sequence;
    对所述信号序列进行滤波，得到滤波后的频谱信号；Filtering the signal sequence to obtain a filtered spectrum signal;
    对所述频率信号进行倒频谱计算得到梅氏倒频谱数据。The cepstrum calculation is performed on the frequency signal to obtain Meyer's-eye cepstrum data.
  6. 一种配电网局部放电超声波检测系统，其特征在于，包括：An ultrasonic detection system for partial discharge in a distribution network, characterized in that it includes:
    训练模块，用于根据历史配电设备局部放电缺陷的超声波信号、历史配电设备局部放电缺陷的图像和历史配电设备检测结果训练神经网络模型，得到训练好的神经网络模型；所述训练好的神经网络模型包括一层周期神经网络层、一层卷积神经网络层和多层全连接层；The training module is used to train the neural network model according to the ultrasonic signal of the partial discharge defect of the historical power distribution equipment, the image of the partial discharge defect of the historical power distribution equipment and the detection result of the historical power distribution equipment, so as to obtain the trained neural network model; The neural network model includes a periodic neural network layer, a convolutional neural network layer and a multi-layer fully connected layer;
    梅氏倒频谱数据获取模块，用于将待测配电网设备的局部放电缺陷的超声波信号转换成梅氏倒频谱数据；The cepstral data acquisition module is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into the cepstrum data;
    第一特征获取模块，用于将所述梅氏倒频谱数据输入所述周期神经网络层进行学习得到第一特征；a first feature acquisition module, configured to input the cepstrum data into the periodic neural network layer for learning to obtain a first feature;
    第二特征获取模块，用于将待测配电网设备的局部放电缺陷的图像输入所述卷积神经网络层进行学习得到第二特征；The second feature acquisition module is used to input the image of the partial discharge defect of the distribution network equipment to be tested into the convolutional neural network layer for learning to obtain the second feature;
    第三特征获取模块，用于将所述第一特征和所述第二特征进行线性拼接后 得到第三特征；The third feature acquisition module is used to obtain the third feature after the first feature and the second feature are linearly spliced;
    检测模块，用于将所述第三特征输入所述多层全连接层，得到待测配电网设备的检测结果。The detection module is used for inputting the third feature into the multi-layer fully connected layer to obtain the detection result of the distribution network equipment to be tested.
  7. 根据权利要求6所述的配电网局部放电超声波检测系统，其特征在于，所述训练模块包括：The ultrasonic detection system for partial discharge in power distribution network according to claim 6, wherein the training module comprises:
    输入单元，用于将所述历史配电设备局部放电缺陷的超声波信号和所述历史配电设备局部放电缺陷的图像输入至所述神经网络模型，得到输出结果；an input unit, configured to input the ultrasonic signal of the partial discharge defect of the historical power distribution equipment and the image of the partial discharge defect of the historical power distribution equipment into the neural network model to obtain an output result;
    调整单元，用于根据所述输出结果以及所述历史配电设备检测结果调整所述神经网络模型，得到训练好的神经网络模型。An adjustment unit, configured to adjust the neural network model according to the output result and the detection result of the historical power distribution equipment to obtain a trained neural network model.
  8. 根据权利要求6所述的配电网局部放电超声波检测系统，其特征在于，所述调整单元包括：The ultrasonic detection system for partial discharge in a distribution network according to claim 6, wherein the adjustment unit comprises:
    判断子单元，用于判断所述输出结果与所述历史配电设备检测结果的误差是否在误差范围内；a judging sub-unit for judging whether the error between the output result and the historical power distribution equipment detection result is within the error range;
    若是，则确定所述神经网络模型为训练好的神经网络模型；If so, then determine that the neural network model is a trained neural network model;
    若否，则利用梯度反向传播算法调整所述神经网络模型，使所述输出结果与所述历史配电设备检测结果的误差在误差范围内。If not, the neural network model is adjusted by using the gradient back-propagation algorithm, so that the error between the output result and the detection result of the historical power distribution equipment is within the error range.
  9. 根据权利要求6所述的配电网局部放电超声波检测系统，其特征在于，所述检测模块得到的所述待测配电网设备的检测结果包括配电设备局部放电的缺陷类型、配电设备故障的严重程度和配电设备的使用寿命。The ultrasonic detection system for partial discharge in a distribution network according to claim 6, wherein the detection result of the distribution network equipment to be tested obtained by the detection module includes the defect type of the partial discharge of the power distribution equipment, the power distribution equipment The severity of the failure and the service life of the power distribution equipment.
  10. 根据权利要求6所述的配电网局部放电超声波检测系统，其特征在于，所述梅氏倒频谱数据获取模块包括：The ultrasonic detection system for partial discharge in a distribution network according to claim 6, wherein the cepstrum data acquisition module comprises:
    数字信号转换单元，用于将待测配电网设备的局部放电缺陷的超声波信号转换成数字信号；The digital signal conversion unit is used to convert the ultrasonic signal of the partial discharge defect of the distribution network equipment to be tested into a digital signal;
    加窗数字信号获取单元，用于对所述数字信号进行加窗处理获得加窗后的数字信号；a windowed digital signal acquisition unit, configured to perform windowing processing on the digital signal to obtain a windowed digital signal;
    信号序列获取单元，用于将所述加窗后的数字信号进行傅里叶变换，得到 信号序列；a signal sequence acquisition unit, for performing Fourier transform on the digital signal after the windowing to obtain a signal sequence;
    频谱信号获取单元，用于对所述信号序列进行滤波，得到滤波后的频谱信号；a spectral signal acquisition unit, configured to filter the signal sequence to obtain a filtered spectral signal;
    梅氏倒频谱数据获取单元，用于对所述频率信号进行倒频谱计算得到梅氏倒频谱数据。The cepstrum data acquisition unit is configured to perform cepstrum calculation on the frequency signal to obtain cepstrum data.
PCT/CN2021/087208 2020-06-28 2021-04-14 Power distribution network partial discharge ultrasonic test method and system WO2022001289A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
ZA2021/04616A ZA202104616B (en) 2020-06-28 2021-07-02 Ultrasonic detection method and system for partial discharge of power distribution network

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010596194.1 2020-06-28
CN202010596194.1A CN111932493B (en) 2020-06-28 2020-06-28 Power distribution network partial discharge ultrasonic detection method and system

Publications (1)

Publication Number Publication Date
WO2022001289A1 true WO2022001289A1 (en) 2022-01-06

Family

ID=73317827

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/087208 WO2022001289A1 (en) 2020-06-28 2021-04-14 Power distribution network partial discharge ultrasonic test method and system

Country Status (4)

Country Link
CN (1) CN111932493B (en)
LU (1) LU102837B1 (en)
WO (1) WO2022001289A1 (en)
ZA (1) ZA202104616B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111932493B (en) * 2020-06-28 2024-06-07 北京国网富达科技发展有限责任公司 Power distribution network partial discharge ultrasonic detection method and system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140372052A1 (en) * 2011-09-19 2014-12-18 Korea Electric Power Corporation Method and device for determining the defect type of a partial discharge
CN106556781A (en) * 2016-11-10 2017-04-05 华乘电气科技（上海）股份有限公司 Shelf depreciation defect image diagnostic method and system based on deep learning
CN109856517A (en) * 2019-03-29 2019-06-07 国家电网有限公司 A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data
CN110162799A (en) * 2018-11-28 2019-08-23 腾讯科技（深圳）有限公司 Model training method, machine translation method and relevant apparatus and equipment
CN111047598A (en) * 2019-11-04 2020-04-21 华北电力大学（保定） Deep learning-based ultraviolet discharge light spot segmentation method and device for power transmission and transformation equipment
CN111932493A (en) * 2020-06-28 2020-11-13 北京国网富达科技发展有限责任公司 Power distribution network partial discharge ultrasonic detection method and system

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106546892A (en) * 2016-11-10 2017-03-29 华乘电气科技（上海）股份有限公司 The recognition methodss of shelf depreciation ultrasonic audio and system based on deep learning

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140372052A1 (en) * 2011-09-19 2014-12-18 Korea Electric Power Corporation Method and device for determining the defect type of a partial discharge
CN106556781A (en) * 2016-11-10 2017-04-05 华乘电气科技（上海）股份有限公司 Shelf depreciation defect image diagnostic method and system based on deep learning
CN110162799A (en) * 2018-11-28 2019-08-23 腾讯科技（深圳）有限公司 Model training method, machine translation method and relevant apparatus and equipment
CN109856517A (en) * 2019-03-29 2019-06-07 国家电网有限公司 A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data
CN111047598A (en) * 2019-11-04 2020-04-21 华北电力大学（保定） Deep learning-based ultraviolet discharge light spot segmentation method and device for power transmission and transformation equipment
CN111932493A (en) * 2020-06-28 2020-11-13 北京国网富达科技发展有限责任公司 Power distribution network partial discharge ultrasonic detection method and system

Also Published As

Publication number Publication date
CN111932493A (en) 2020-11-13
ZA202104616B (en) 2022-07-27
LU102837B1 (en) 2022-01-03
CN111932493B (en) 2024-06-07

Similar Documents

Publication Publication Date Title
CN111929542B (en) Power equipment diagnosis method and system
WO2021107342A1 (en) Device and method for monitoring state of vibration of rotating equipment using deep learning-based time series analysis
US20230015731A1 (en) Real-time abnormity-diagnosing and interpolation method for water regime-monitoring data
CN109029582A (en) A kind of Power Transformer Faults on-line detecting system based on many kinds of parameters acquisition
CN116320832B (en) Monitoring equipment fault monitoring method and device
WO2022001289A1 (en) Power distribution network partial discharge ultrasonic test method and system
CN111239672B (en) Machine learning algorithm-based gradient fault prediction method for optical fiber current transformer
WO2023279382A1 (en) Motor bearing running state fault detection method and system
CN103235993A (en) Electric energy quality evaluation system and method for enterprise user side
CN113298134A (en) BPNN-based remote non-contact health monitoring system and method for fan blade
CN115528810A (en) Temperature measurement data operation, inspection, analysis and management system based on power equipment
CN111679176A (en) Circuit fault monitoring system for power saving
CN114157023B (en) Distribution transformer early warning information acquisition method
CN116823226A (en) Electric power district fault monitoring system based on big data
CN109285331B (en) Power cable temperature early warning system based on data analysis and temperature prediction
CN208399633U (en) High-voltage cable GIS terminal Partial Discharge Detection and diagnostic system
CN202351373U (en) Power cable fault traveling wave distance measuring equipment
CN116561707B (en) Transformer fault checking and early warning method and system
CN113109666A (en) Track circuit fault diagnosis method based on deep convolutional neural network
CN111456915A (en) Fault diagnosis device and method for internal components of fan engine room
CN205941719U (en) Removal operation management end that power equipment detected
FR3099830B1 (en) Method and system for monitoring a cable network, by principal component analysis
CN109581109A (en) A kind of state of elevator multi-information fusion diagnostic method
CN114089136A (en) High-voltage cable partial discharge detection and defect diagnosis system based on artificial intelligence
CN113688791A (en) Wind turbine CMS abnormal data packet identification method

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: 21833813

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: 21833813

Country of ref document: EP

Kind code of ref document: A1