CN113221688B - Knife switch state identification method, device and storage medium - Google Patents
Knife switch state identification method, device and storage medium Download PDFInfo
- Publication number
- CN113221688B CN113221688B CN202110462824.0A CN202110462824A CN113221688B CN 113221688 B CN113221688 B CN 113221688B CN 202110462824 A CN202110462824 A CN 202110462824A CN 113221688 B CN113221688 B CN 113221688B
- Authority
- CN
- China
- Prior art keywords
- training
- disconnecting link
- image
- model
- sample set
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000003860 storage Methods 0.000 title claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 95
- 238000001514 detection method Methods 0.000 claims abstract description 50
- 238000012360 testing method Methods 0.000 claims description 26
- 238000011156 evaluation Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 13
- 230000009466 transformation Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000003321 amplification Effects 0.000 claims description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 4
- 238000012790 confirmation Methods 0.000 abstract description 5
- 238000004422 calculation algorithm Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 10
- 238000013135 deep learning Methods 0.000 description 6
- 238000002372 labelling Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 230000011664 signaling Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 238000011426 transformation method Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- 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
Abstract
The invention discloses a knife switch state identification method, a device and a storage medium, wherein the method comprises the following steps: acquiring a disconnecting link image of a state to be identified; inputting the acquired disconnecting link image into a pre-trained disconnecting link state identification model, and determining the opening and closing states of the disconnecting links according to the output of the disconnecting link state identification model; the knife switch state recognition model adopts a CenterNet target detection network, training samples of the knife switch state recognition model are images of multiple types of knife switches of a transformer substation with known knife switch state types in different states, and output of the knife switch state recognition model comprises opening and closing state information and type information of the knife switches in the input image. The disconnecting link state identification method disclosed by the invention can meet the two-position confirmation requirement of the disconnecting link of the transformer substation in a one-key sequential control application scene, and improves the efficiency of detecting the disconnecting link state of the transformer substation.
Description
Technical Field
The invention relates to the technical field of operation and maintenance of transformer substations, in particular to a method, a device and a storage medium for identifying the state of a transformer substation disconnecting link based on deep learning.
Background
The state detection of the isolating switch of the transformer substation is an important ring in the operation and maintenance of the transformer substation. The separation switch is normally checked by the separation and closing contact signals of an auxiliary switch or manually. In the application scene of one-key sequential control, the state position identification of the disconnecting link is very critical, and if the auxiliary switch switching signal is only relied on to serve as the criterion of the one-key sequential control, the reliability is relatively low. Therefore, the related technical specifications provide that the isolating knife switch state must be confirmed by two positions of 'position remote signaling + non-homologous remote signaling', the 'position remote signaling' can be ensured by an auxiliary contact signal, and the traditional 'non-homologous remote signaling' generally adopts a manual visual inspection mode, so that the efficiency is low, the power failure time is long, and the intelligent degree is low.
Noun interpretation
CenterNet: a classical Anchor-Free target detection algorithm can directly detect the center point and the size of a target, and the detection speed and the detection precision are good.
mAP: mean Average Precision average accuracy, i.e. average accuracy first within one class (Average Precision), then average accuracy again for all classes (mean Average Precision). Is a performance evaluation index of the target detection model.
Libtach is a basic component of the deep learning typical framework pytorch that is developed to be open to the C++ environment deployment of the model.
Disclosure of Invention
The invention aims to provide a substation disconnecting link state identification method and device, which can meet the two-position confirmation requirement of a substation disconnecting link in a one-key sequential control application scene and improve the efficiency of substation disconnecting link state detection. The technical scheme adopted by the invention is as follows.
In one aspect, the invention provides a method for identifying a state of a knife switch, which comprises the following steps:
acquiring a disconnecting link image of a state to be identified;
inputting the acquired disconnecting link image into a pre-trained disconnecting link state identification model, and determining the opening and closing states of the disconnecting links according to the output of the disconnecting link state identification model;
the training method of the disconnecting link state recognition model comprises the following steps:
obtaining image samples of various types of disconnecting links of the transformer substation in different states, and obtaining an initial sample library;
preprocessing an image sample in an initial sample library to obtain a training sample library;
acquiring manual annotation information of each image sample in a training sample library, and establishing a training sample set and a test sample set;
training a pre-constructed CenterNet target detection network by using a training sample set and a test sample set to obtain a CenterNet target detection model.
Optionally, in the training method of the disconnecting link state identification model, the preprocessing includes: classifying pattern samples according to the type, the mounting mode and the switching state of the knife switch in the image;
sample expansion is carried out on pattern samples under each category through random copying and/or data enhancement processing, so that training samples with balanced sample numbers under each category are obtained, and a training sample library is formed.
Optionally, in the training method of the disconnecting link state recognition model, the preprocessing further includes, for the difficult-to-recognize image samples with complex disconnecting link background, increasing the number of sample expansion in the training sample library. Therefore, the generalization capability of the model obtained by final training can be improved, and the model has better recognition capability for the state of the disconnecting link with complex background.
Optionally, the data enhancement process includes performing a geometric transformation process and/or a color transformation process on the sample image, the geometric transformation process including one or more of flipping, rotating, clipping, deforming, scaling, and the color transformation process including one or more of noise, blurring, color transformation, erasure, filling.
Optionally, in the training method of the disconnecting link state identification model, the obtaining of the manual labeling information of the image sample includes:
after a key identification area of a disconnecting link state in an image sample is marked by a labelImg tool manually, an xml format marking file which comprises marking area coordinate information and state classification information is generated;
and converting the xml format annotation file into the json format annotation file. The state classification information is classification type information of the disconnecting link and disconnecting link opening and closing state information, so that the model can be used for reasoning according to the input image to obtain the state classification information of the disconnecting link in the image in the application after training is completed.
Optionally, the training method of the knife switch state identification model further comprises the following steps: converting the CenterNet target detection model into a torchscript format model file supported by a libtorch component package;
in the knife switch state identification method, a C++ interface provided by a libtorch component is used for calling a centrinet target detection model in a torchscript format, and the obtained knife switch image is used as the input of the model to obtain the output information of the model.
The model conversion algorithm refers to a conversion algorithm provided by the pytorch and capable of converting the pytorch model into a c++ interface readable model, and can be in a trace or script mode. The libtorch component package is a basic component which is pushed out by pyrtorch and is opened for the C++ environment deployment of the model, and the main application scene of torchScript is to convert PyTorch codes into equivalent C++ codes, so that the operation efficiency of a deep learning model can be improved.
Optionally, in the training method of the disconnecting link state recognition model, training the pre-built centrnet target detection network by using a training sample set and a test sample set includes:
a) Training the CenterNet target detection network by using a training sample set to obtain an intermediate model file;
b) Using mAP@0.5 as a model evaluation index, evaluating the intermediate model file by using a test sample set, and judging whether the evaluation index meets a preset requirement:
if the central net target detection model is reached, stopping training to obtain the central net target detection model after training;
if not, carrying out amplification and/or data enhancement processing on the image samples of the current training sample library, acquiring new manual annotation information, establishing a new training sample set and a new testing sample set, and transferring to the steps a) -b) based on the new training sample set and the new testing sample set until the evaluation index of the intermediate model file reaches the preset requirement, and stopping training.
In a second aspect, the present invention provides a knife switch state identifying device, including:
the disconnecting link image acquisition module to be identified is configured to acquire a disconnecting link image of a state to be identified;
the disconnecting link state recognition module is configured to input the acquired disconnecting link image into a pre-trained disconnecting link state recognition model, and determine the opening and closing state of the disconnecting link according to the output of the disconnecting link state recognition model;
the method comprises the steps that a central Net target detection network is adopted in a disconnecting link state identification model, training samples of the central Net target detection network are images of a plurality of types of disconnecting links of a transformer substation with known disconnecting link state types in different states, and output of the disconnecting link state identification model comprises opening and closing state information and type information of the disconnecting links in an input image.
In a third aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the knife switch state identification method according to the first aspect.
Advantageous effects
The invention realizes the identification of the knife switch state based on the CenterNet target detection algorithm, does not need to manually set the anchor frame during application, has strong real-time performance, small calculated amount and high identification accuracy, and can meet the requirements of one-key sequential control on two-position confirmation and the like.
Meanwhile, the invention converts the CenterNet target detection model into the torchscript format model file supported by the libtorch component package, and calls the CenterNet target detection model through the C++ interface provided by the libtorch component to realize the reasoning of the knife switch state in the image to be identified, thereby obtaining better running efficiency in a computer and further improving the real-time performance of the knife switch state identification.
Drawings
FIG. 1 is a schematic flow chart of a knife switch state recognition method according to an embodiment of the present invention;
fig. 2 is a schematic diagram of an application interface of a one-key sequential control background using the knife switch state recognition method according to the present invention in an application example.
Detailed Description
Further description is provided below in connection with the drawings and the specific embodiments.
Example 1
The embodiment introduces a method for identifying the state of a knife switch, which comprises the following steps:
acquiring a disconnecting link image of a state to be identified;
inputting the acquired disconnecting link image into a pre-trained disconnecting link state identification model, and determining the opening and closing states of the disconnecting links according to the output of the disconnecting link state identification model;
the training method of the disconnecting link state recognition model comprises the following steps:
obtaining image samples of various types of disconnecting links of the transformer substation in different states, and obtaining an initial sample library;
preprocessing an image sample in an initial sample library to obtain a training sample library;
acquiring manual annotation information of each image sample in a training sample library, and establishing a training sample set and a test sample set;
training a pre-constructed CenterNet target detection network by using a training sample set and a test sample set to obtain a CenterNet target detection model.
Referring to fig. 1, the following specifically describes what the present embodiment relates to.
1. Training of knife switch state recognition model
1.1 training sample acquisition
Firstly, collecting image samples of various types of disconnecting link of a transformer substation, and establishing an initial sample library. Preferably, the images of the disconnecting link on and off states under different angles, different illumination conditions and the like are collected, all types of on and off state materials need to be completely covered, and the number of samples exceeds 5000.
And classifying images in the initial sample library according to different appearances/mounting modes and different opening and closing states of the knife switch, and expanding few classified samples by a data enhancement technology to obtain a sample library with balanced classification. The knife gate image is classified into 500heng_open, 500heng_close, 500shu_open, 500shu_close, 220heng_open, 220heng_close, 220shu_open, 220shu_close, 35_open, and 35_close.
For samples with less classification, a method of randomly copying the classified image samples and then carrying out data enhancement is adopted, so that the number of the classified samples is ensured to be approximately consistent. The data enhancement method comprises geometrical transformation methods such as turning, rotating, cutting, deforming, scaling and the like, and color transformation methods such as noise, blurring, color transformation, erasing, filling and the like. The processing method for the single image sample may be any one or a combination of plural kinds thereof.
In addition, aiming at difficult-to-identify samples with complex backgrounds, the expansion quantity of the samples is increased, and the generalization capability of a model obtained through final training can be improved.
After the sample image is obtained, the key identification area of the image in the sample library can be marked manually, the computer obtains the manual marking information to form an xml format marking file containing the coordinate and classification information, and then the xml format marking file is converted into a json format marking file supported by a central Net algorithm through a python script.
In this embodiment, an open-source labelmg tool may be used to label the sample library image, where the labeling area is a key identification area of the knife switch, and a classification name to which the area belongs is set, so as to form an xml format labeling file. After the sample labeling is completed, a training sample set and a test sample set are established according to the proportion, and the proportion is recommended to be 4:1.
1.2 model training
The current deep learning methods in the field of target detection are mainly divided into two categories: a target detection algorithm of the two stage; a one stage target detection algorithm. The former is that a series of candidate frames serving as samples are generated by an algorithm, and then the samples are classified by a convolutional neural network; the latter directly converts the problem of target frame positioning into regression problem processing without generating candidate frames. The difference of the two methods is also reflected in the performance, the two stage is superior in detection accuracy and positioning accuracy, and the one stage is superior in algorithm speed. The CenterNet algorithm is one of the best methods for performing (and improving) the one-stage target detection method in practical tests. The method changes the target detection problem into a standard key point estimation problem, transmits the image into a full convolution network to obtain a thermodynamic diagram, wherein the peak point of the thermodynamic diagram is a central point, the peak point position of each characteristic diagram predicts the width and height information of the target, and finally outputs a final prediction result according to the confidence coefficient.
Therefore, in this embodiment, a central net target detection network training knife switch state recognition model is selected, as shown in fig. 1, and the method includes the following steps:
a) Training the CenterNet target detection network by using a training sample set to obtain an intermediate model file; the intermediate model file stores information such as a deep learning convolutional network structure, weight parameters of each layer and the like;
b) Using mAP@0.5 as a model evaluation index, evaluating the intermediate model file by using a test sample set, and judging whether the evaluation index meets a preset requirement:
if the central net target detection model is reached, stopping training to obtain the central net target detection model after training;
if not, carrying out amplification and/or data enhancement processing on the image samples of the current training sample library, acquiring new manual annotation information, establishing a new training sample set and a new testing sample set, and transferring to the steps a) -b) based on the new training sample set and the new testing sample set until the evaluation index of the intermediate model file reaches the preset requirement, and stopping training.
If the evaluation index does not meet the technical specification requirements, the training and testing sample set needs to be reestablished in a mode of optimizing sample labeling and continuing to amplify samples. The optimized sample labeling method may be an adjustment or subdivision class. The amplified samples may be more new image samples or may be duplicated on an existing sample basis for data enhancement. And (5) obtaining an optimal model file through repeated iterative training and evaluation.
1.3 model File conversion
In this embodiment, after model training is completed, the optimal model file is converted into an optimal model file in the torchscript format supported by the libtorch component package through the deep learning framework pytorch. The libtorch component package is a basic component which is developed by PyTorch and aims at C++ environment deployment of a model, and the main application scene of torchScript is to convert PyTorch codes into equivalent C++ codes, so that the operation efficiency of a deep learning model is improved. The pyrach code is compiled into a format that the torchscript compiler can understand, a file in this format can be loaded in the c++ encoding environment and the corresponding code executed with built-in JIT.
2. Knife switch state identification
When the trained disconnecting link identification model is used for actual disconnecting link state identification, disconnecting link image data to be identified are firstly obtained. The method comprises the steps of encoding and decoding video data by using an encoder based on knife switch real-time video data acquired by a plurality of fixed point cameras, and acquiring decoded video data. The image file of the state of the knife switch to be detected can be captured from the video data periodically in the process of executing the one-key sequential control operation.
And then, based on a C++ reasoning interface provided by the libtorch component, taking all images to be detected as input, and obtaining a product file containing the on-off state and the classification type of the disconnecting link through the reasoning process of the optimal model file in the torchscript format. And the monitoring background reads and analyzes the product file in real time, so that the state information of all the disconnecting links of the current transformer substation can be obtained, the judgment of the states of the disconnecting links based on intelligent image analysis is realized, and the double-confirmation requirement of one-key sequential control operation is met.
Example 2
This embodiment introduces a switch state recognition device, includes:
the disconnecting link image acquisition module to be identified is configured to acquire a disconnecting link image of a state to be identified;
the disconnecting link state recognition module is configured to input the acquired disconnecting link image into a pre-trained disconnecting link state recognition model, and determine the opening and closing state of the disconnecting link according to the output of the disconnecting link state recognition model;
the method comprises the steps that a central Net target detection network is adopted in a disconnecting link state identification model, training samples of the central Net target detection network are images of a plurality of types of disconnecting links of a transformer substation with known disconnecting link state types in different states, and output of the disconnecting link state identification model comprises opening and closing state information and type information of the disconnecting links in an input image.
The training process of the knife switch state recognition model and the realization of the image processing and recognition functions in practical application are specifically referred to embodiment 1.
Example 3
This embodiment describes a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the knife switch state identification method described in embodiment 1.
In summary, the method and the device can realize the judgment of the state of the knife switch based on intelligent image analysis and meet the double-confirmation requirement of one-key sequential control operation. The one-key sequential control background main interface based on the intelligent analysis of the deep learning image can be shown by referring to fig. 2.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The embodiments of the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present invention and the scope of the claims, which are all within the protection of the present invention.
Claims (5)
1. A method for identifying the state of a knife switch is characterized by comprising the following steps:
acquiring a disconnecting link image of a state to be identified;
inputting the acquired disconnecting link image into a pre-trained disconnecting link state identification model, and determining the opening and closing states of the disconnecting links according to the output of the disconnecting link state identification model;
the training method of the disconnecting link state recognition model comprises the following steps:
obtaining image samples of various types of disconnecting links of the transformer substation in different states, and obtaining an initial sample library;
preprocessing an image sample in an initial sample library to obtain a training sample library;
acquiring manual annotation information of each image sample in a training sample library, and establishing a training sample set and a test sample set;
training a pre-constructed CenterNet target detection network by using a training sample set and a test sample set to obtain a CenterNet target detection model;
in the training method of the disconnecting link state recognition model, the preprocessing comprises the following steps: classifying image samples according to the type, the mounting mode and the switching state of the knife switch in the image;
sample expansion is carried out on the image samples under each category through random copying and/or data enhancement processing, so that training samples with balanced sample numbers under each category are obtained, and a training sample library is formed; for the difficult-to-identify image samples with complex knife switch backgrounds, increasing the sample expansion quantity of the difficult-to-identify image samples in a training sample library;
the obtaining of the manual annotation information of the image sample comprises the following steps:
after a key identification area of a disconnecting link state in an image sample is marked by a labelImg tool manually, an xml format marking file which comprises marking area coordinate information and state classification information is generated;
converting the xml format annotation file into a json format annotation file;
the training of the pre-constructed CenterNet target detection network by using the training sample set and the test sample set comprises the following steps:
a) Training the CenterNet target detection network by using a training sample set to obtain an intermediate model file;
b) Using mAP@0.5 as a model evaluation index, evaluating the intermediate model file by using a test sample set, and judging whether the evaluation index meets a preset requirement:
if the central net target detection model is reached, stopping training to obtain the central net target detection model after training;
if not, carrying out amplification and/or data enhancement processing on the image samples of the current training sample library, acquiring new manual annotation information, establishing a new training sample set and a new testing sample set, and transferring to the steps a) -b) based on the new training sample set and the new testing sample set until the evaluation index of the intermediate model file reaches the preset requirement, and stopping training.
2. The method of claim 1, wherein the data enhancement process comprises performing a geometric transformation process and/or a color transformation process on the sample image, the geometric transformation process comprising one or more of flipping, rotating, cropping, morphing, scaling, and the color transformation process comprising one or more of noise, blurring, color transformation, erasure, padding.
3. The method of claim 1, wherein the training method of the knife switch state recognition model further comprises: converting the CenterNet target detection model into a torchscript format model file supported by a libtorch component package;
in the knife switch state identification method, a C++ interface provided by a libtorch component is used for calling a centrinet target detection model in a torchscript format, and the obtained knife switch image is used as the input of the model to obtain the output information of the model.
4. A knife switch state recognition device is characterized by comprising:
the disconnecting link image acquisition module to be identified is configured to acquire a disconnecting link image of a state to be identified;
the disconnecting link state recognition module is configured to input the acquired disconnecting link image into a pre-trained disconnecting link state recognition model, and determine the opening and closing state of the disconnecting link according to the output of the disconnecting link state recognition model;
the method comprises the steps that a central Net target detection network is adopted in a disconnecting link state identification model, training samples of the central Net target detection network are images of a plurality of types of disconnecting links of a transformer substation with known disconnecting link state types in different states, and output of the disconnecting link state identification model comprises opening and closing state information and type information of the disconnecting links in an input image;
the training method of the disconnecting link state recognition model comprises the following steps:
obtaining image samples of various types of disconnecting links of the transformer substation in different states, and obtaining an initial sample library;
preprocessing an image sample in an initial sample library to obtain a training sample library;
acquiring manual annotation information of each image sample in a training sample library, and establishing a training sample set and a test sample set;
training a pre-constructed CenterNet target detection network by using a training sample set and a test sample set to obtain a CenterNet target detection model;
in the training method of the disconnecting link state recognition model, the preprocessing comprises the following steps: classifying image samples according to the type, the mounting mode and the switching state of the knife switch in the image;
sample expansion is carried out on the image samples under each category through random copying and/or data enhancement processing, so that training samples with balanced sample numbers under each category are obtained, and a training sample library is formed; for the difficult-to-identify image samples with complex knife switch backgrounds, increasing the sample expansion quantity of the difficult-to-identify image samples in a training sample library;
the obtaining of the manual annotation information of the image sample comprises the following steps:
after a key identification area of a disconnecting link state in an image sample is marked by a labelImg tool manually, an xml format marking file which comprises marking area coordinate information and state classification information is generated;
converting the xml format annotation file into a json format annotation file;
the training of the pre-constructed CenterNet target detection network by using the training sample set and the test sample set comprises the following steps:
a) Training the CenterNet target detection network by using a training sample set to obtain an intermediate model file;
b) Using mAP@0.5 as a model evaluation index, evaluating the intermediate model file by using a test sample set, and judging whether the evaluation index meets a preset requirement:
if the central net target detection model is reached, stopping training to obtain the central net target detection model after training;
if not, carrying out amplification and/or data enhancement processing on the image samples of the current training sample library, acquiring new manual annotation information, establishing a new training sample set and a new testing sample set, and transferring to the steps a) -b) based on the new training sample set and the new testing sample set until the evaluation index of the intermediate model file reaches the preset requirement, and stopping training.
5. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method of identifying a knife switch state according to any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110462824.0A CN113221688B (en) | 2021-04-28 | 2021-04-28 | Knife switch state identification method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110462824.0A CN113221688B (en) | 2021-04-28 | 2021-04-28 | Knife switch state identification method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113221688A CN113221688A (en) | 2021-08-06 |
| CN113221688B true CN113221688B (en) | 2024-03-19 |
Family
ID=77089317
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110462824.0A Active CN113221688B (en) | 2021-04-28 | 2021-04-28 | Knife switch state identification method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113221688B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113792795A (en) * | 2021-09-15 | 2021-12-14 | 南方电网数字电网研究院有限公司 | Substation equipment identification method and device based on gateway equipment and gateway equipment |
| CN113569819B (en) * | 2021-09-23 | 2021-12-24 | 泰豪软件股份有限公司 | Method and system for identifying opening and closing states of disconnecting link, storage medium and computer equipment |
| CN114049598B (en) * | 2022-01-13 | 2022-04-19 | 国网浙江省电力有限公司宁波供电公司 | State identification method and device of power primitive, storage medium and electronic equipment |
| CN114202731A (en) * | 2022-02-15 | 2022-03-18 | 南京天创电子技术有限公司 | Multi-state knob switch identification method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046943A (en) * | 2019-12-09 | 2020-04-21 | 国网智能科技股份有限公司 | Method and system for automatically identifying state of isolation switch of transformer substation |
| CN111626987A (en) * | 2020-04-29 | 2020-09-04 | 南京邮电大学 | Vehicle detection method based on unmanned aerial vehicle and anchor-frame-free network |
| CN111652873A (en) * | 2020-06-04 | 2020-09-11 | 天津工业大学 | Permanent magnet self-unloading iron remover iron-discarding conveyer belt fracture detection method based on transfer learning |
| CN111814742A (en) * | 2020-07-29 | 2020-10-23 | 南方电网数字电网研究院有限公司 | Knife switch state identification method based on deep learning |
| CN112418024A (en) * | 2020-11-10 | 2021-02-26 | 北京五八信息技术有限公司 | Target identification method and device, mobile terminal and storage medium |
-
2021
- 2021-04-28 CN CN202110462824.0A patent/CN113221688B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046943A (en) * | 2019-12-09 | 2020-04-21 | 国网智能科技股份有限公司 | Method and system for automatically identifying state of isolation switch of transformer substation |
| CN111626987A (en) * | 2020-04-29 | 2020-09-04 | 南京邮电大学 | Vehicle detection method based on unmanned aerial vehicle and anchor-frame-free network |
| CN111652873A (en) * | 2020-06-04 | 2020-09-11 | 天津工业大学 | Permanent magnet self-unloading iron remover iron-discarding conveyer belt fracture detection method based on transfer learning |
| CN111814742A (en) * | 2020-07-29 | 2020-10-23 | 南方电网数字电网研究院有限公司 | Knife switch state identification method based on deep learning |
| CN112418024A (en) * | 2020-11-10 | 2021-02-26 | 北京五八信息技术有限公司 | Target identification method and device, mobile terminal and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113221688A (en) | 2021-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113221688B (en) | Knife switch state identification method, device and storage medium | |
| CN112131936B (en) | Inspection robot image recognition method and inspection robot | |
| CN112633231B (en) | Fire disaster identification method and device | |
| CN110969130A (en) | Driver dangerous action identification method and system based on YOLOV3 | |
| CN101986348A (en) | Visual target identification and tracking method | |
| CN110378221A (en) | A kind of power grid wire clamp detects and defect identification method and device automatically | |
| CN110598693A (en) | Ship plate identification method based on fast-RCNN | |
| CN110795990A (en) | Gesture recognition method for underwater equipment | |
| CN112528979B (en) | Transformer substation inspection robot obstacle distinguishing method and system | |
| CN113313037A (en) | Method for detecting video abnormity of generation countermeasure network based on self-attention mechanism | |
| CN113642474A (en) | Hazardous area personnel monitoring method based on YOLOV5 | |
| CN106339657A (en) | Straw incineration monitoring method and device based on monitoring video | |
| CN114898466A (en) | Video motion recognition method and system for smart factory | |
| CN111145222A (en) | Fire detection method combining smoke movement trend and textural features | |
| WO2021147055A1 (en) | Systems and methods for video anomaly detection using multi-scale image frame prediction network | |
| CN113792578A (en) | Method, device and system for detecting abnormity of transformer substation | |
| Manninen et al. | Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images | |
| WO2020008459A1 (en) | System and method for use in object detection from video stream | |
| CN111563886B (en) | Unsupervised feature learning-based tunnel steel rail surface disease detection method and device | |
| CN115861948A (en) | Lane line detection method, lane line detection device, lane line early warning method, lane line early warning system and medium | |
| CN114998580A (en) | Fracture semantic segmentation method based on deep LabV3+ network model | |
| CN113743306A (en) | Method for analyzing abnormal behaviors of real-time intelligent video monitoring based on slowfast double-frame rate | |
| CN113378598A (en) | Dynamic bar code detection method based on deep learning | |
| CN112132088A (en) | Inspection point location missing inspection identification method | |
| CN117315354B (en) | Insulator anomaly detection method based on multi-discriminant composite coding GAN 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |