CN114519293A - Cable body fault identification method based on hand sample machine learning model - Google Patents
Cable body fault identification method based on hand sample machine learning model Download PDFInfo
- Publication number
- CN114519293A CN114519293A CN202111611821.5A CN202111611821A CN114519293A CN 114519293 A CN114519293 A CN 114519293A CN 202111611821 A CN202111611821 A CN 202111611821A CN 114519293 A CN114519293 A CN 114519293A
- Authority
- CN
- China
- Prior art keywords
- machine learning
- learning model
- cable body
- small sample
- method based
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/16—Cables, cable trees or wire harnesses
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- 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
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a cable body fault identification method based on a small sample machine learning model, which comprises the following steps: s1, collecting induced current signals of different faults of the power cable, and establishing a sample library; s2, preprocessing the sample library in the S1, and constructing a small sample training set, a verification set and a test set; s3, constructing a small sample machine learning model based on a convolutional neural network and metric learning; s4, training the machine learning model by the training set constructed in the S2; and S5, periodically verifying the machine learning model by the verification set constructed in the S2, and solidifying the model when the verification training reaches a set accuracy rate. And S6, testing the model, inputting the power cable induction current harmonic signals to the curing model in S5, outputting corresponding classification labels, and finally obtaining the category of the power cable body fault. The method can effectively avoid the problems of difficult data sample collection, large labeling workload and the like in engineering application, and is particularly suitable for the condition of few data samples in the engineering application.
Description
Technical Field
The invention relates to a cable body fault identification method based on a small sample machine learning model. Belongs to the technical field of power cable body fault diagnosis.
Background
The safety of the power cable, which is used as a carrier for transmitting electric energy in power supply, is very important. However, due to poor laying environment (cable trenches and bridges) and influence of an early construction process (unreasonable sealing of cable heads and terminal heads, scratch of an outer sheath and the like), the cable is easy to be damaged or the insulation is aged to cause failure. With the increase of the operation time, the failure rate is higher and higher, and once a failure occurs, the failure diagnosis is relatively difficult, the repair difficulty is high, the time consumption is long, and the great loss is inevitable to be solved.
With the development of technology and the continuous improvement of computing power, the classification and recognition method based on artificial intelligence makes great progress in many aspects such as images, voice and the like, and makes great contribution to the global economic growth. As a class of complex machine learning algorithms with huge parameters, the learning effect needs a large amount of labeled data support.
More and more scenes can not provide sufficient samples for learning, for example, in the aspect of fingerprint identification of communication radiation sources, under the actual complex electromagnetic environment conditions, for each electromagnetic environment radiation source, people can hardly acquire sufficient radiation source observation sample data of known types, and the traditional machine learning classification identification algorithm is limited in use in such cases.
The traditional machine learning classification and identification algorithm needs a huge data system for support, needs to consume larger manpower and physics, and has low model training speed and lower operation flexibility.
Disclosure of Invention
The invention aims to overcome the defects and provides a cable body fault identification method based on a small sample machine learning model.
The purpose of the invention is realized as follows:
a cable body fault identification method based on a small sample machine learning model is characterized by comprising the following steps: the method comprises the following steps:
s1, collecting induced current signals of different faults of the power cable, and establishing a sample library;
s2, preprocessing the sample library in the S1, and constructing a small sample training set, a verification set and a test set;
s3, constructing a small sample machine learning model based on a convolutional neural network and metric learning;
s4, training the machine learning model by the training set constructed in the S2;
and S5, regularly verifying the machine learning model by the verification set constructed in the S2, and solidifying the model when the verification training reaches the set accuracy.
And S6, testing the model, inputting the power cable induction current harmonic signals to the curing model in S5, outputting corresponding classification labels, and finally obtaining the category of the power cable body fault.
After the training of the learning model of the machine of the hand sample is finished, the induced current harmonic signals of the power cable are input in real time, the trained model parameters are loaded, and the fault category can be output in real time.
Further, the data preprocessing comprises data cleaning, data extraction, data labeling and feature tag construction; s2 includes the steps of:
s2-1, data cleaning: deleting invalid data and repeated data in the original sample library established in the step S1, and processing missing values and abnormal values;
s2-2, data extraction: fourier transformation is carried out on the induced current signals in the sample library processed by the S2-1, and current harmonic signals of 2-10 times are extracted;
s2-3, data annotation: labeling the sample obtained in the step S2-2 with a label, wherein the label is a cable body fault category;
s2-4, constructing a feature label.
Further, the data extraction in S2-2 may also employ wavelet transform.
Further, the structural feature labels comprise a total harmonic distortion rate, a higher harmonic ratio and a harmonic content rate.
Further, the data set comprises 800 classes, 500 classes are randomly selected as a training set, 200 classes are selected as a verification set, and the remaining 100 classes are selected as a test set; the categories of the training set, validation set, and test set do not intersect.
Further, in S1, the cable induced current signal includes a current harmonic signal of 2 to 10 times of the induced current, time information, and identification information of the cable and category information of the fault.
Further, in S3, the convolutional neural network is a feature descriptor for signal detection and identification, and includes a convolutional layer, a pooling layer, and a batch normalization layer.
Further, the induced current signal was characterized using Conv4, each volume block consisting of a convolution layer of 64 3 × 3 convolution kernels, a batch normalization layer, a Relu activation function layer, and a 2 × 2 max pooling layer.
Further, in S3, the metric means of metric learning includes manhattan distance, euclidean distance, mahalanobis distance, chebyshev distance, and cosine distance.
Further, in S4, the training set data of S2 is input into the small sample machine learning model, N classes are randomly extracted from the training set every training period, k samples of each class are taken as a support set S, b samples are taken as a query set Q, and the objective function is as follows:
wherein, S and B are respectively a support sample set and a query sample set which are randomly extracted from a training set, theta is a parameter set of the model, x represents a sample in the query sample set B, and y represents a prediction label of x.
Furthermore, the cosine distance is adopted to calculate the similarity between samples when the vector a (x)11 , x12 , x13, …, x1n) And b (x)21,x22,x23,…,x2n) All in one vector space, then there are:
normalizing the cosine distance, wherein a = softmax (cos θ), and the prediction types of the input samples are as follows:
wherein the content of the first and second substances,afor the trained attention kernel, k is the number of classes of the dataset M,x irepresenting the marked samples in the data set M,y irepresentx iThe label of (1).
Compared with the prior art, the invention has the beneficial effects that:
the method for identifying the fault of the cable body based on the small sample machine learning model can effectively avoid the difficulties that data samples are difficult to collect, the labeling workload is large and the like in engineering application, has high efficiency and practicability, and is particularly suitable for scenes which are limited by environment and cannot provide sufficient samples for learning.
According to the cable body fault identification method based on the small sample machine learning model, training can be completed only by a relatively small number of samples, the defect that a training model is required to be marked by a large number of samples in the traditional machine learning method is overcome, high instantaneity and high efficiency are achieved, a large amount of manpower and material resources can be saved, low-cost and high-flexibility operation is achieved, and the obtained verification result represents that the accuracy is higher than 85%.
The invention relates to a cable body fault identification method based on a small sample machine learning model, which is particularly suitable for the condition of few data samples in engineering application.
Drawings
Fig. 1 is a flowchart of a power cable body fault identification method of a cable body fault identification method based on a hand sample machine learning model according to the present invention.
FIG. 2 is a flow chart of a model building algorithm.
Fig. 3 is a diagram of a backbone network structure.
FIG. 4 is a schematic diagram of a model training process.
Detailed Description
The following description of the embodiments of the present invention will be made with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
Referring to fig. 1, the embodiment provides a cable body fault identification method based on a small sample machine learning model, including the following steps:
s1, collecting data;
collecting induced current signals of different power cable body faults, including but not limited to 2-10 times of current harmonic signals and time information of the induced current, identification information of the cable, category information of the faults, other additional information and the like, numbering the signals respectively, and establishing a sample library;
s2, preprocessing the data of the sample library, and constructing a training set, a verification set and a test set;
the data preprocessing comprises data cleaning, data extraction, data labeling and feature tag construction;
the purpose of data cleaning is to delete invalid data and repeated data in the original sample library and process missing values and abnormal values.
The data extraction includes but is not limited to fourier transform and wavelet transform, mainly extracts 2-10 times of current harmonic signals from induced current signals of the cable, and in the embodiment, fourier transform extraction is adopted.
The data labeling is to label the extracted data and classify the data according to the same registration number, wherein the label comprises the following fault categories: the fault of the bubble of the cable insulating layer, the fault of the water branch of the insulating layer, the fault of the bulge of the insulating layer and the fault of the bulge of the inner semi-conducting layer caused by the impurity contained in the insulating layer.
The structural characteristic labels comprise characteristic indexes such as total harmonic distortion rate, higher harmonic ratio, harmonic content rate and the like.
Dividing a data set, and constructing a training set, a verification set and a test set, wherein the data set after pretreatment in the embodiment comprises 800 classes, 500 classes are randomly selected as the training set, 200 classes are selected as the verification set, and the remaining 100 classes are selected as the test set; the categories of the training set, validation set, and test set do not intersect.
S3, constructing a small sample machine learning model based on a convolutional neural network and metric learning;
the small sample machine learning model consists of a backbone network and a similarity measure, and the algorithm flow is shown in FIG. 2. Referring to fig. 3, the backbone network uses a convolutional neural network as a feature descriptor, the convolutional neural network commonly used includes, but is not limited to, Vggl6, residual error networks Resnet, inclusion, a feature pyramid network FPN, and the like, in this embodiment, Conv4 is used to perform feature extraction on the induced current signal, each convolutional block is composed of a convolutional layer of 64 convolutional kernels 3 × 3, a batch normalization layer, a Relu activation function layer, and a 2 × 2 max pooling layer, and an output result is a feature vector of the signal mapped to a high-dimensional space.
The metric learning is to learn a nearest classifier for making the similarity of the same type samples large and the similarity of the different type samples small; mapping the samples to a high-dimensional metric space through a convolutional neural network, and measuring the similarity between the samples in the high-dimensional metric space; the measurement means includes Manhattan distance, Euclidean distance, Mahalanobis distance, and tangentThe cosine distance is used to calculate the similarity between samples in this embodiment as the vector a (x)11 , x12 , x13, …, x1n) And b (x)21,x22,x23,…,x2n) All in one vector space, then there are:
normalizing the cosine distance, wherein a = softmax (cos θ), and the prediction types of the input samples are as follows:
wherein, the first and the second end of the pipe are connected with each other,afor the trained attention kernel, k is the number of classes of the dataset M,x irepresenting the marked samples in the data set M,y irepresentx iThe label of (1).
S4, training the machine learning model by the training set constructed in the S2;
inputting the training set data in S2 into a small sample machine learning model, as shown in fig. 4, in each training period, randomly extracting N classes from the training set, sampling k samples in each class as a support set S, sampling B samples as a query set Q, and maximizing the probability that the support set S predicts the labels in the query set B, where the objective function is:
s and B are respectively a support sample set and a query sample set which are randomly extracted from a training set, theta is a parameter set of the model, x represents a sample in the query sample set B, and y represents a prediction label of x.
S5, verifying and training;
and (5) periodically verifying the machine learning model by using the verification set constructed in the S2, and solidifying the model after the verification training reaches the set accuracy.
S6, test model, inputting power cable induction current harmonic wave signal to the solidification model in S5 Output the corresponding classification label And finally obtaining the category of the power cable body fault.
After the training of the learning model of the machine of the hand sample is finished, the induced current harmonic signals of the power cable are input in real time, the trained model parameters are loaded, and the fault category can be output in real time.
Compared with the traditional machine learning method which needs a large number of marked samples to train the model, the method only needs to mark a small number of samples, is favorable for realizing low cost, has high real-time performance and high efficiency, and can save a large amount of manpower and material resources.
In the above embodiments, the present invention is described only by way of example, but those skilled in the art, after reading the present patent application, may make various modifications to the present invention without departing from the spirit and scope of the present invention.
Claims (11)
1. A cable body fault identification method based on a small sample machine learning model is characterized in that: the method comprises the following steps:
s1, collecting induced current signals of different faults of the power cable, and establishing a sample library;
s2, preprocessing the sample library in the S1, and constructing a small sample training set, a verification set and a test set;
s3, constructing a small sample machine learning model based on a convolutional neural network and metric learning;
s4, training the machine learning model by the training set constructed in the S2;
and S5, regularly verifying the machine learning model by the verification set constructed in the S2, and solidifying the model when the verification training reaches the set accuracy.
And S6, testing the model, inputting the power cable induction current harmonic signals to the curing model in S5, outputting corresponding classification labels, and finally obtaining the category of the power cable body fault.
2. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, wherein: the data preprocessing comprises data cleaning, data extraction, data labeling and feature tag construction; s2 includes the steps of:
s2-1, data cleaning: deleting invalid data and repeated data in the original sample library established in the step S1, and processing missing values and abnormal values;
s2-2, data extraction: fourier transformation is carried out on the induced current signals in the sample library processed by the S2-1, and current harmonic signals of 2-10 times are extracted;
s2-3, data annotation: labeling the sample obtained in the step S2-2 with a label, wherein the label is a cable body fault category;
s2-4, constructing a feature label.
3. The cable body fault identification method based on the small sample machine learning model as claimed in claim 2, characterized in that: the data extraction in S2-2 may also employ wavelet transform.
4. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, wherein: the construction characteristic labels comprise a total harmonic distortion rate, a higher harmonic ratio and a harmonic content rate.
5. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, wherein: the data set comprises 800 classes, 500 classes are randomly selected as a training set, 200 classes are selected as a verification set, and the rest 100 classes are selected as a test set; the categories of the training set, validation set, and test set do not intersect.
6. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, wherein: in S1, the cable induced current signal includes a current harmonic signal of 2 to 10 times of the induced current, time information, and identification information of the cable and type information of the fault.
7. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, wherein: in S3, the convolutional neural network is a feature descriptor for signal detection and identification, and includes a convolutional layer, a pooling layer, and a batch normalization layer.
8. The cable body fault identification method based on the small sample machine learning model as claimed in claim 7, wherein: induced current signals were characterized using Conv4, each convolution block consisting of a convolution layer of 64 3 x 3 convolution kernels, a batch normalization layer, a Relu activation function layer, and a 2 x 2 max pooling layer.
9. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, characterized in that: in S3, the metric means for metric learning includes manhattan distance, euclidean distance, mahalanobis distance, chebyshev distance, and cosine distance.
10. The cable body fault identification method based on the small sample machine learning model as claimed in claim 1, characterized in that: in S4, the training set data of S2 is input into a small sample machine learning model, N classes are randomly extracted from the training set every training period, k samples of each class are used as a support set S, b samples are used as a query set Q, and the objective function is:
wherein, S and B are respectively a support sample set and a query sample set which are randomly extracted from a training set, theta is a parameter set of the model, x represents a sample in the query sample set B, and y represents a prediction label of x.
11. The cable body fault identification method based on the small sample machine learning model as claimed in claim 9, wherein: calculating the similarity between samples by using cosine distance as vector a (x)11 , x12 , x13, …, x1n) And b (x)21,x22,x23,…,x2n) All in one vector space, then there are:
normalizing the cosine distance, wherein a = softmax (cos θ), and the prediction types of the input samples are as follows:
wherein the content of the first and second substances,afor the trained attention kernel, k is the number of classes of the dataset M,x irepresenting the marked samples in the data set M,y ito representx iThe label of (1).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111611821.5A CN114519293A (en) | 2021-12-27 | 2021-12-27 | Cable body fault identification method based on hand sample machine learning model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111611821.5A CN114519293A (en) | 2021-12-27 | 2021-12-27 | Cable body fault identification method based on hand sample machine learning model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114519293A true CN114519293A (en) | 2022-05-20 |
Family
ID=81596741
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111611821.5A Pending CN114519293A (en) | 2021-12-27 | 2021-12-27 | Cable body fault identification method based on hand sample machine learning model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114519293A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115879049A (en) * | 2023-03-02 | 2023-03-31 | 国网江西省电力有限公司电力科学研究院 | Induction identification output method and system of automatic vertical rod |
CN117520818A (en) * | 2023-11-08 | 2024-02-06 | 广州水沐青华科技有限公司 | Power fingerprint identification method and device, electronic equipment and storage medium |
-
2021
- 2021-12-27 CN CN202111611821.5A patent/CN114519293A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115879049A (en) * | 2023-03-02 | 2023-03-31 | 国网江西省电力有限公司电力科学研究院 | Induction identification output method and system of automatic vertical rod |
CN117520818A (en) * | 2023-11-08 | 2024-02-06 | 广州水沐青华科技有限公司 | Power fingerprint identification method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110879989B (en) | Ads-b signal target identification method based on small sample local machine learning model | |
CN111337768B (en) | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil | |
CN110943857B (en) | Power communication network fault analysis and positioning method based on convolutional neural network | |
CN109635928A (en) | A kind of voltage sag reason recognition methods based on deep learning Model Fusion | |
CN114519293A (en) | Cable body fault identification method based on hand sample machine learning model | |
CN112101085B (en) | Intelligent fault diagnosis method based on importance weighted domain antagonism self-adaptation | |
CN105468677A (en) | Log clustering method based on graph structure | |
CN113592822B (en) | Insulator defect positioning method for electric power inspection image | |
CN115994325B (en) | Fan icing power generation data enhancement method based on TimeGAN deep learning method | |
CN110675421B (en) | Depth image collaborative segmentation method based on few labeling frames | |
CN111008641B (en) | Power transmission line tower external force damage detection method based on convolutional neural network | |
CN112367273A (en) | Knowledge distillation-based flow classification method and device for deep neural network model | |
CN112419268A (en) | Method, device, equipment and medium for detecting image defects of power transmission line | |
CN111242028A (en) | Remote sensing image ground object segmentation method based on U-Net | |
CN117272230A (en) | Non-invasive load monitoring method and system based on multi-task learning model | |
CN112507881A (en) | sEMG signal classification method and system based on time convolution neural network | |
CN116611580A (en) | Ocean red tide prediction method based on multi-source data and deep learning | |
CN116503612A (en) | Fan blade damage identification method and system based on multitasking association | |
CN116205905A (en) | Power distribution network construction safety and quality image detection method and system based on mobile terminal | |
CN110705638A (en) | Credit rating prediction classification method using deep network learning fuzzy information feature technology | |
CN114519294A (en) | Power cable fault identification method based on wavelet singular entropy and BP neural network | |
Shaotong et al. | Location and identification of insulator and bushing based on YOLOv3-spp algorithm | |
CN112016403B (en) | Video abnormal event detection method | |
CN114841266A (en) | Voltage sag identification method based on triple prototype network under small sample | |
CN113705695A (en) | Power distribution network fault data identification method based on convolutional neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |