CN110175571A - The intellectual monitoring of substation equipment state and recognition methods - Google Patents
The intellectual monitoring of substation equipment state and recognition methods Download PDFInfo
- Publication number
- CN110175571A CN110175571A CN201910451544.2A CN201910451544A CN110175571A CN 110175571 A CN110175571 A CN 110175571A CN 201910451544 A CN201910451544 A CN 201910451544A CN 110175571 A CN110175571 A CN 110175571A
- Authority
- CN
- China
- Prior art keywords
- image
- image set
- insulator
- network model
- subgraph
- 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
Classifications
-
- 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
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of intellectual monitoring of substation equipment state and recognition methods, comprising the following steps: the video or image information of acquisition power transformation station equipment, to establish the original data collection of equipment；Original data collection is divided into disconnecting switch image set, line map image set and insulator image set according to the classification of equipment；The sample image in each image set is pre-processed respectively, to obtain disconnecting switch subgraph image set, route subgraph image set and insulator subgraph image set；It is respectively corresponded according to each subgraph image set and establishes disconnecting switch confrontation network model, route confrontation network model and insulator confrontation network model；Disconnecting switch, route and the insulator in test image are monitored and are identified respectively using each confrontation network model.The intellectual monitoring of the substation equipment state and recognition methods are able to solve the status monitoring problem of large-scale equipment, and improve recognition accuracy while reducing handmarking's amount.
Description
Technical field
The present invention relates to identification technology field more particularly to a kind of intellectual monitoring of substation equipment state and identification sides
Method.
Background technique
In recent years, State Grid Corporation of China increases the supervision to substation safety level run, the work of overhaul of the equipments
It measures and increases therewith, but traditional maintenance mode based on preventive trial is there are heavy workload, process is complicated, safety coefficient is low
The problems such as.With the fast development of economic society and the continuous enlargement of power grid scale, user is to the reliability of system power supply and steady
It is qualitative that more stringent requirements are proposed.The intellectual monitoring of status of electric power can with the operating status of real-time tracking equipment, in time on
Faulty equipment information is passed, the division position of disconnecting switch is such as identified, monitors the ageing state etc. of insulator.Therefore, it reliably sets
Standby Condition Monitoring Technology can be effectively reduced the influence of power outage, guarantee that power grid is continually and steadily run.
Currently, the technology for electrical equipment status monitoring is mainly sensor technology and image recognition technology, sensor
Technology includes the technologies such as pressure sensor, attitude transducer and acceleration transducer, is chiefly used in sentencing for disconnecting switch division position
Not, but the switchgear of not applicable part specific model, and there is a problem of taking electric difficult, signal transmission poor.With sensor skill
Art is compared, and image recognition technology is widely used, both can be with identification switch division shape suitable for multiple types or the equipment of pattern
State, and can be with monitoring device failure, recognition accuracy is high, and recognition result is reliable.However, existing image recognition technology is being set
Standby monitoring field using generally existing following problem, first is that image processing techniques mainly uses edge extracting, straight-line detection and
The methods of sign location, this method manual labor amount is big, and efficiency is lower, is unable to Real-Time Sharing data resource；Second is that not having
Handle the ability of the equipment image pattern of magnanimity；Third is that time-consuming for identification process, low efficiency leads to data transfer delay, information
Sharing is poor, influences operation of power networks stability.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.For this purpose, of the invention
One purpose is to propose intellectual monitoring and the recognition methods of a kind of substation equipment state, to solve the shape of large-scale equipment
State monitoring problem, and handmarking's amount is reduced, improve recognition accuracy.
In order to achieve the above objectives, the invention proposes a kind of intellectual monitoring of substation equipment state and recognition methods, packets
Include following steps: the video or image information of acquisition power transformation station equipment, to establish the original data collection of equipment, wherein
The equipment includes disconnecting switch, route, insulator；According to the classification of equipment by the original data collection be divided into every
It leaves and closes image set, line map image set and insulator image set；Respectively to the disconnecting switch image set, the circuit image
Sample image in collection and the insulator image set is pre-processed, to obtain disconnecting switch subgraph image set, route subgraph
Collection and insulator subgraph image set；According to the disconnecting switch subgraph image set, the route subgraph image set and the insulator subgraph
Image set, which respectively corresponds, establishes disconnecting switch confrontation network model, route confrontation network model and insulator confrontation network model；Benefit
With disconnecting switch confrontation network model, route confrontation network model and insulator confrontation network model difference
Disconnecting switch, route and insulator in test image is monitored and is identified.
The intellectual monitoring of the substation equipment state of the embodiment of the present invention and recognition methods are made by fighting study mechanism
The sample that generator generates has more diversity, plays the role of data enhancing, model is made to be not easy to fall into part in the training process
Minimum point is able to solve the status monitoring problem of large-scale equipment, and it is quasi- to improve identification while reducing handmarking's amount
True rate, and can be realized the normal and abnormality of substation isolating-switch division state and malfunction, route and insulator
On-line intelligence monitoring, help to improve system run all right.
Wherein, the video of the power transformation station equipment or image information be one of in the following manner or a variety of acquisitions:
Video detecting device, unmanned plane technology, the route inspecting robot being mounted in substation acquire.
Wherein, using Scale invariant features transform method according to the classification of equipment by the original data collection be divided into every
It leaves and closes image set, line map image set and insulator image set.
Specifically, described respectively to the disconnecting switch image set, the line map image set and the insulator image set
In sample image pre-processed, comprising: by gray scale operation by the sample image in each image set by triple channel image turn
Become single channel image；Each single channel image is stored in the form of two dimensional data structure；Pass through one-dimensional wavelet function point
It is other that every a line of each single channel image and each column are converted, obtain the approximation component and details point of two-dimensional wavelet transformation
Amount, wherein the approximation component includes Two-order approximation component, and the details coefficients include that second order level detail component, second order hang down
Straight details coefficients, second order diagonal detail component, single order level detail component, single order vertical detail component and single order diagonal detail
Component；Using the Two-order approximation component of each single channel image as the corresponding subgraph of each sample image.
Wherein, gray scale operation is carried out by following formula:
F=0.30R+0.59G+0.11B,
Wherein, F is the gray value of single channel image, and R, G, B are respectively pixel value of the triple channel image on each channel.
Specifically, described according to the disconnecting switch subgraph image set, the route subgraph image set and the insulator subgraph
Image set, which respectively corresponds, establishes disconnecting switch confrontation network model, route confrontation network model and insulator confrontation network model, packet
It includes: concentrating the size of image to establish the corresponding initial confrontation network model of each equipment respectively according to each subgraph；Utilize each subgraph
Image set, and the corresponding training sample image of each equipment and its status indication value in material database are called, to each initial confrontation network
Model is trained to obtain each confrontation network model, wherein the training sample image includes the equipment under a variety of weather environments
Image.
Wherein, the corresponding initial confrontation network model of the disconnecting switch is 5 layer networks, and the route and insulator are corresponding
Initial confrontation network model be 3 layer networks, and 5 layer network and 3 layer network include generator and differentiation
Device, and network final layer uses sigmoid activation primitive, remainder layer is all made of leaky_relu activation primitive.
Specifically, the objective function L of the generatorGIt is indicated by following formula:
Wherein, f (x) indicates the middle layer output of arbiter；
The objective function L of the arbiterDIt is indicated by following formula:
LD=Lsupevised+Lunsupevised,
Wherein,Indicate supervised learning loss,Indicate without
Supervised learning loss, Plabeled(x,y)Middle x indicates the training sample image in material database, and y indicates the corresponding status indication value of x,
Punlabeled(x)Middle x indicates the sample image that each subgraph is concentrated, and z indicates the input noise of the generator, k presentation class number,
The "false" image of generator output is classified as+1 class of kth.
Wherein, the status indication value of the disconnecting switch include be closed completely, be fully disconnected, transition state and failure shape
The status indication value of state, the route and the insulator includes normal condition and malfunction.
Wherein, a variety of weather environments include sunlight, sleet, haze and night.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Fig. 1 is the flow chart of the intellectual monitoring and recognition methods of the substation equipment state of the embodiment of the present invention；
Fig. 2 is the flow chart of the intellectual monitoring and recognition methods of the substation equipment state of a specific example of the invention；
Fig. 3 is the wavelet decomposition subband figure of an example of the present invention；
Fig. 4 is the structural schematic diagram of the generator of an example of the present invention；
Fig. 5 is the structural schematic diagram of the arbiter of an example of the present invention；
Fig. 6 is test sample-test result curve graph of an example of the present invention.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to
The embodiment of attached drawing description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
Below with reference to the accompanying drawings intellectual monitoring and the recognition methods of the substation equipment state of the embodiment of the present invention are described.
Fig. 1 is the flow chart of the intellectual monitoring and recognition methods of the substation equipment state of the embodiment of the present invention.
As shown in Figure 1, the intellectual monitoring of substation equipment state and recognition methods the following steps are included:
S1 acquires the video or image information of power transformation station equipment, to establish the original data collection of equipment, wherein
Equipment includes disconnecting switch, route, insulator.
In this embodiment, the video of power transformation station equipment or image information can be one of in the following manner or a variety of
Acquisition: video detecting device, unmanned plane technology, the route inspecting robot being mounted in substation acquire.
It specifically, can be by the video detecting device, unmanned plane or the route inspecting robot that are mounted in substation
Deng the video or image information of acquisition equipment, and collected video or image information can be stored in monitoring center database
In.When needing to be monitored equipment and identify, image pattern and corresponding can be extracted by information database first out of slave station
Device status information is respectively formed original data collection and state array, and state array includes that can recognize to set in correspondence image
Standby all fault messages or location information.Wherein, original data collection and state array can be carried out according to number in station
Arrangement, convenient for the realization of the method for the present invention.
S2, according to the classification of equipment by original data collection be divided into disconnecting switch image set, line map image set and absolutely
Edge subgraph image set.
In this embodiment, Scale invariant features transform method can be used to be drawn original data collection according to the classification of equipment
It is divided into disconnecting switch image set, line map image set and insulator image set.
Specifically, as shown in Figure 1, there are plurality of devices since original data is concentrated, it is therefore desirable to be set to different
It is standby, mainly image-region positioning, identification and the extraction of disconnecting switch, route and insulator equipment.Scale can be used in the present invention
Invariant features convert (Scale-invariant feature transform, SIFT) algorithm to the equipment in initial pictures into
Row Region Matching.SIFT algorithm is a kind of operator of description image local feature based on scale space, the algorithm distinction
Good, informative, local feature also keeps certain stability to visual angle change, affine transformation, noise.The algorithm flow
It is as follows,
1) scale space is constructed, extreme point is detected, obtains scale invariability；
2) it is accurately positioned key point Location Scale, removes unstable edge respective point；
3) each key point is described, generate characteristic matching vector, wherein SIFT feature vector to illumination, noise,
Rotation and scale have good invariance；
4) Image Feature Matching and extracted region of target device.
Disconnecting switch, route and the insulation subgraph of characteristic feature be will be provided with as matching stencil, incited somebody to action by SIFT method
The equipment that original data is concentrated carries out images match with template respectively, marks and extracts the rectangle region where each equipment
Domain divides one group into a kind of equipment image is belonged to, and forms the initial subgraph image set of multiple groups, the sample of every group of subgraph image set according to
Initial pictures sequential organization in step S1, and guarantee that the size of image pattern in same subgraph image set is identical.
S3 respectively locates the sample image in disconnecting switch image set, line map image set and insulator image set in advance
Reason, to obtain disconnecting switch subgraph image set, route subgraph image set and insulator subgraph image set.
In this embodiment, respectively to the sample graph in disconnecting switch image set, line map image set and insulator image set
As being pre-processed, comprising: operated by gray scale the sample image in each image set being changed into single channel by triple channel image
Image；Each single channel image is stored in the form of two dimensional data structure；By one-dimensional wavelet function respectively to each single-pass
Every a line of road image and each column are converted, and the approximation component and details coefficients of two-dimensional wavelet transformation are obtained, wherein close
It include Two-order approximation component like component, details coefficients include second order level detail component, second order vertical detail component, second order pair
Angle details coefficients, single order level detail component, single order vertical detail component and single order diagonal detail component；By each single channel figure
The Two-order approximation component of picture is as the corresponding subgraph of each sample image.Wherein, gray scale operation can be carried out by following formula:
F=0.30R+0.59G+0.11B (1)
Wherein, F is the gray value of single channel image, and R, G, B are respectively pixel value of the triple channel image on each channel.
Specifically, the sample in each image set still includes a large amount of environmental information and disturbing factor.Preprocessing process can
The irrelevant information amount and data dimension of a picture is effectively reduced, the processing time is reduced, recognition efficiency is improved and identification is accurate
Rate.Initial color image includes R, G, B Three-channel data value in computer systems, is operated first by gray scale by triple channel
Image is changed into single channel image, determines gray value F using weighted mean method, as shown in above formula (1).
Image after gray processing is stored in a computer in the form of two dimensional data structure, is distinguished by one-dimensional wavelet function
Its every a line and each column are converted, the approximation component and details coefficients of two-dimensional wavelet transformation are obtained, approximation component indicates
The low frequency value of signal, details coefficients indicate the high frequency value of signal.As shown in figure 3, cA2, cH2, cV2, cD2 respectively indicate image
Two-order approximation component, level detail component, vertical detail component and diagonal detail component, cH1, cV1, cD1 respectively indicate one
Rank level, vertical and diagonal detail component.Second order low-frequency approximation component in subband contains the overwhelming majority in initial pictures
Information, and only have 1/16 size of original image in size, the feature letter of equipment is also remained while reducing image data amount
Breath.The subgraph image set of each group of equipment obtained in step S2 is pre-processed by the above method, extracts Two-order approximation point
Amount forms input data matrix of the new subgraph image set as following model.
S4 respectively corresponds foundation isolation according to disconnecting switch subgraph image set, route subgraph image set and insulator subgraph image set
Switch confrontation network model, route confrontation network model and insulator fight network model.
In this embodiment, right respectively according to disconnecting switch subgraph image set, route subgraph image set and insulator subgraph image set
Disconnecting switch confrontation network model, route confrontation network model and insulator confrontation network model should be established, comprising: according to each
Subgraph concentrates the size of image to establish the corresponding initial confrontation network model of each equipment respectively；Using each subgraph image set, and adjust
With the corresponding training sample image of each equipment and its status indication value in material database, each initial confrontation network model is carried out
Training obtains each confrontation network model, wherein training sample image includes the equipment image under a variety of weather environments.Wherein, every
The status indication value for leaving pass include be closed completely, be fully disconnected, transition state and malfunction, the shape of route and insulator
State mark value includes normal condition and malfunction.
Optionally, a variety of weather environments include sunlight, sleet, haze and night.
Wherein, the corresponding initial confrontation network model of disconnecting switch is 5 layer networks, and route and insulator are corresponding initial right
Anti- network model is 3 layer networks, and 5 layer networks and 3 layer networks include generator and arbiter, and network final layer is adopted
With sigmoid activation primitive, remainder layer is all made of leaky_relu activation primitive.Wherein, the objective function L of generatorGPass through
Following formula (2) indicates:
Wherein, f (x) indicates the middle layer output of arbiter, connects the output of generator gradually by characteristic matching mechanism
Nearly true picture.
The objective function L of arbiterDIt is indicated by following formula (2):
LD=Lsupevised+Lunsupevised (3)
Wherein,Indicate supervised learning loss,Indicate without
Supervised learning loss, Plabeled(x,y)Middle x indicates the training sample image in material database, and y indicates the corresponding status indication value of x,
Punlabeled(x)Middle x indicates that the sample image that each subgraph is concentrated, z indicate the input noise of generator, and k presentation class number is raw
The "false" image of output of growing up to be a useful person is classified as+1 class of kth.
Specifically, as shown in Fig. 2, pretreated subgraph image set is respectively disconnecting switch subgraph image set, insulator subgraph
Image set and route subgraph image set, image pattern therein are the Two-order approximation component extracted after wavelet transformation.According to every
The sample-size of a sub- image set determines optimal network hyper parameter respectively, establishes three groups of confrontation network models later and recognizes respectively
The state of disconnecting switch, insulator and route.
1) disconnecting switch recognizes
Before and after grid switching operation, usually there are three kinds of states in disconnecting switch: close a floodgate completely, complete separating brake, transition state.This hair
The network model of bright middle foundation can realize the state recognition to disconnecting switch state according to image pattern, and can identify
It has the disconnecting switch there are failure.To i-th of disconnecting switch QiHave:
Wherein, tiThe status indication value for indicating i-th of disconnecting switch, using one group of four-dimensional one-hot coding vector pair
Answer tiFour kinds of state classifications, and according to the chronological order of disconnecting switch image data set formed status indication encoder matrix.It is right
In a certain group of disconnecting switch (A/B/C three-phase), when only exporting 01 and 10, control system can just assert that switch is in and arrive completely
Position state, otherwise output is waited instruction or fault message to be uploaded to mster-control centre by system.
For the sample-size of disconnecting switch image data set, the generator and arbiter of 5 layer networks are established.Generator
Structure is as shown in figure 4, the effect of generator is that random noise input is reduced to true picture size step by step.Arbiter knot
Structure and generator structure are on the contrary, as shown in Figure 5.By calling disconnecting switch training image sample and its state in material database
Mark value completes model training, i.e., the sample matrix of disconnecting switch subgraph image set obtained above and calls from material database
Disconnecting switch training image sample and its status indication value are passed through as authentic specimen with the "false" sample matrix that generator generates
Fourth dimension forms joint input matrix after being coupled and inputs as arbiter, by two layers of convolutional network and one layer of fully-connected network
Exporting one group of five dimension one-hot vector afterwards indicates the classification results of state.The joint input square of this group of output vector and arbiter
Sample in battle array corresponds, four kinds of state+generator "false" image classifications of five kinds of classification corresponding (4).
The arbiter objective function L of the semi-supervised trained mechanism as a result,DAs shown in above formula (3), generator objective function LG
As shown in above formula (2).
2) route and insulator state recognize
The state of route and insulator is divided into normal condition and malfunction.Route and insulator image pattern are built respectively
Found 3 layers of confrontation network model, same route and insulator training image sample and its status indication value instruction using in material database
Practice model, arbiter classification results are in addition to the two states in following formula (5), including the third "false" image classification.
S5 is distinguished using disconnecting switch confrontation network model, route confrontation network model and insulator confrontation network model
Disconnecting switch, route and insulator in test image is monitored and is identified.
Wherein, test image is preferably the image that above-mentioned each subgraph is concentrated.
Specifically, realization can be programmed in MATLAB according to process shown in Fig. 2, and right respectively using Neural Network Toolbox
Disconnecting switch, route and insulator sample set are tested, and recognition accuracy is recorded.Test parameter is provided that for isolation
Switch samples collection establishes 5 layers of confrontation network, establishes 3 layers of confrontation network respectively for route and insulation submodel.Network final layer
Using sigmoid activation primitive, remainder layer is all made of leaky_relu activation primitive.To prevent over-fitting, arbiter
Coefficient 0.5 is arranged in dropout, and tape label sample (sample called from material database) measures accounting 0.2.Used test sample
Information is as shown in table 1,
Table 1
It is right using the present invention based on GAN (Generative Adversarial Networks, production fight network)
Every group of sample is tested in table 1, and be based on CNN (Convolutional Neural Networks, convolutional Neural net
Network) image recognition algorithm comparison, I Classification and Identification accuracy rate of gained test group is listed in table 2, test result, and test is whole accurate
Rate is listed in table 3.
Table 2
| Recognition methods | Disconnecting switch | Route | Insulator | Whole accuracy rate |
| GAN | 95.83% | 100% | 100% | 96.67% |
| CNN+SVM | 93.75% | 100% | 83.33% | 93.33% |
Table 3
| Test group | Ⅰ | Ⅱ | Ⅲ | Ⅳ |
| GAN | 96.67% | 96.21% | 91.99% | 93.04% |
| CNN+SVM | 93.33% | 89.01% | 76.17% | 71.57% |
From Table 2, it can be seen that discrimination of the method for the present invention on sorting device is superior to CNN algorithm, from Fig. 6
Out, with the growth of sample size, advantage of this paper model in large-scale image identification problem is more obvious.
The intellectual monitoring of the substation equipment state of the embodiment of the present invention and recognition methods are made by fighting study mechanism
The sample that generator generates has more diversity, plays the role of data enhancing, model is made to be not easy to fall into part in the training process
Minimum point；The status monitoring that large-scale equipment is solved the problems, such as using semi-supervised learning mechanism is reducing the same of handmarking's amount
When improve recognition accuracy；By region recognition and extract to device class, according to classification results adjust automatically network inputs
Hyper parameter is exported to adapt to the size of distinct device sample, improves operation efficiency, shortens the model running time, data is reduced and passes
Defeated delay can be realized the normal and abnormal shape of substation isolating-switch division state and malfunction, route and insulator
The on-line intelligence of state monitors, and helps to improve system run all right.
Further, the invention also provides a kind of computer readable storage mediums.
In an embodiment of the present invention, computer program, computer program quilt are stored on computer readable storage medium
Processor realizes intellectual monitoring and the recognition methods of above-mentioned substation equipment state when executing.
The computer readable storage medium of the embodiment of the present invention, the intelligence with above-mentioned substation equipment state stored on it
When energy monitoring computer program corresponding with recognition methods is executed by processor, substation isolating-switch division shape can be realized
The on-line intelligence of the normal and abnormality of state and malfunction, route and insulator monitors, and it is steady to help to improve system operation
It is qualitative, and recognition accuracy is improved while reducing handmarking's amount.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. intellectual monitoring and the recognition methods of a kind of substation equipment state, which comprises the following steps:
The video or image information for acquiring power transformation station equipment, to establish the original data collection of equipment, wherein the equipment
Including disconnecting switch, route, insulator；
The original data collection is divided into disconnecting switch image set, line map image set and insulator according to the classification of equipment
Image set；
The sample image in the disconnecting switch image set, the line map image set and the insulator image set is carried out respectively
Pretreatment, to obtain disconnecting switch subgraph image set, route subgraph image set and insulator subgraph image set；
Foundation is respectively corresponded according to the disconnecting switch subgraph image set, the route subgraph image set and the insulator subgraph image set
Disconnecting switch fights network model, route confrontation network model and insulator and fights network model；
Network model is fought using disconnecting switch confrontation network model, route confrontation network model and the insulator
Disconnecting switch, route and the insulator in test image are monitored and are identified respectively.
2. intellectual monitoring and the recognition methods of substation equipment state according to claim 1, which is characterized in that the change
The video or image information of electric station equipment be one of in the following manner or a variety of acquisitions:
Video detecting device, unmanned plane technology, the route inspecting robot being mounted in substation acquire.
3. intellectual monitoring and the recognition methods of substation equipment state according to claim 1, which is characterized in that use ruler
It spends invariant features converter technique and the original data collection is divided by disconnecting switch image set, line map according to the classification of equipment
Image set and insulator image set.
4. intellectual monitoring and the recognition methods of substation equipment state according to claim 1, which is characterized in that described point
The other sample image in the disconnecting switch image set, the line map image set and the insulator image set is located in advance
Reason, comprising:
It is operated by gray scale and the sample image in each image set is changed into single channel image by triple channel image；
Each single channel image is stored in the form of two dimensional data structure；
Every a line of each single channel image and each column are converted respectively by one-dimensional wavelet function, obtain 2-d wavelet change
The approximation component and details coefficients changed, wherein the approximation component includes Two-order approximation component, and the details coefficients include second order
Level detail component, second order vertical detail component, second order diagonal detail component, single order level detail component, single order vertical detail
Component and single order diagonal detail component；
Using the Two-order approximation component of each single channel image as the corresponding subgraph of each sample image.
5. intellectual monitoring and the recognition methods of substation equipment state according to claim 4, which is characterized in that by such as
Lower formula carries out gray scale operation:
F=0.30R+0.59G+0.11B,
Wherein, F is the gray value of single channel image, and R, G, B are respectively pixel value of the triple channel image on each channel.
6. intellectual monitoring and the recognition methods of substation equipment state according to claim 1, which is characterized in that described
It respectively corresponds foundation according to the disconnecting switch subgraph image set, the route subgraph image set and the insulator subgraph image set and keeps apart
It closes confrontation network model, route confrontation network model and insulator and fights network model, comprising:
The size of image is concentrated to establish the corresponding initial confrontation network model of each equipment respectively according to each subgraph；
Using each subgraph image set, and the corresponding training sample image of each equipment and its status indication value in material database are called, it is right
Each initial confrontation network model is trained to obtain each confrontation network model, wherein the training sample image includes a variety of days
Equipment image under compression ring border.
7. intellectual monitoring and the recognition methods of substation equipment state according to claim 6, which is characterized in that it is described every
Leaving and closing corresponding initial confrontation network model is 5 layer networks, the route and the corresponding initial confrontation network model of insulator
It is 3 layer networks, and 5 layer network and 3 layer network include generator and arbiter, and network final layer uses
Sigmoid activation primitive, remainder layer are all made of leaky_relu activation primitive.
8. intellectual monitoring and the recognition methods of substation equipment state according to claim 7, which is characterized in that
The objective function L of the generatorGIt is indicated by following formula:
Wherein, f (x) indicates the middle layer output of arbiter；
The objective function L of the arbiterDIt is indicated by following formula:
LD=Lsupevised+Lunsupevised,
Wherein,Indicate supervised learning loss,Indicate without
Supervised learning loss, Plabeled(x,y)Middle x indicates the training sample image in material database, and y indicates the corresponding status indication value of x,
Punlabeled(x)Middle x indicates the sample image that each subgraph is concentrated, and z indicates the input noise of the generator, k presentation class number,
The "false" image of generator output is classified as+1 class of kth.
9. intellectual monitoring and the recognition methods of substation equipment state according to claim 6, which is characterized in that it is described every
The status indication value for leaving pass include be closed completely, be fully disconnected, transition state and malfunction, the route and the insulation
The status indication value of son includes normal condition and malfunction.
10. intellectual monitoring and the recognition methods of substation equipment state according to claim 6, which is characterized in that described
A variety of weather environments include sunlight, sleet, haze and night.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910451544.2A CN110175571A (en) | 2019-05-28 | 2019-05-28 | The intellectual monitoring of substation equipment state and recognition methods |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910451544.2A CN110175571A (en) | 2019-05-28 | 2019-05-28 | The intellectual monitoring of substation equipment state and recognition methods |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110175571A true CN110175571A (en) | 2019-08-27 |
Family
ID=67696433
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910451544.2A Pending CN110175571A (en) | 2019-05-28 | 2019-05-28 | The intellectual monitoring of substation equipment state and recognition methods |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110175571A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942805A (en) * | 2019-12-11 | 2020-03-31 | 云南大学 | Insulator element prediction system based on semi-supervised deep learning |
| CN111027565A (en) * | 2019-12-05 | 2020-04-17 | 国网电力科学研究院武汉南瑞有限责任公司 | Power transmission and transformation equipment report content generation method based on optical vision semantic extraction |
| CN111126417A (en) * | 2020-01-17 | 2020-05-08 | 苏州浪潮智能科技有限公司 | Data center machine room management method and device |
| CN111446724A (en) * | 2020-03-17 | 2020-07-24 | 天津大学 | Multi-source interconnected micro-grid frequency coordination control method based on sliding mode algorithm |
| CN111600386A (en) * | 2020-05-18 | 2020-08-28 | 山东电工电气集团有限公司 | GIS equipment knife switch on-off state confirmation-oriented real-time monitoring system and method |
| CN112396038A (en) * | 2020-07-30 | 2021-02-23 | 国网四川省电力公司信息通信公司 | Transformer substation abnormity intelligent identification method and system based on deep learning |
| CN112836315A (en) * | 2021-02-24 | 2021-05-25 | 上海交通大学 | Neural network-based limit switch production line abnormity monitoring method |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106339722A (en) * | 2016-08-25 | 2017-01-18 | 国网浙江省电力公司杭州供电公司 | Line knife switch state monitoring method and device |
| CN107220929A (en) * | 2017-06-23 | 2017-09-29 | 深圳市唯特视科技有限公司 | A kind of non-paired image method for transformation using the consistent confrontation network of circulation |
| CN108154072A (en) * | 2016-12-02 | 2018-06-12 | 天津工业大学 | Insulator breakdown of taking photo by plane based on depth convolutional neural networks detects automatically |
| CN108171117A (en) * | 2017-12-05 | 2018-06-15 | 南京南瑞信息通信科技有限公司 | Electric power artificial intelligence visual analysis system based on multinuclear heterogeneous Computing |
| CN108389137A (en) * | 2018-02-06 | 2018-08-10 | 国网山西省电力公司电力科学研究院 | Power fault detection early warning system based on infared spectrum technology |
| CN108665058A (en) * | 2018-04-11 | 2018-10-16 | 徐州工程学院 | A kind of generation confrontation network method based on segmentation loss |
| CN108764298A (en) * | 2018-04-29 | 2018-11-06 | 天津大学 | Electric power image-context based on single classifier influences recognition methods |
| CN108828409A (en) * | 2018-08-03 | 2018-11-16 | 南方电网科学研究院有限责任公司 | Fault detection system based on edge calculation |
| CN109117863A (en) * | 2018-07-04 | 2019-01-01 | 全球能源互联网研究院有限公司 | The insulation subsample extending method and device of confrontation network are generated based on depth convolution |
| CN109360188A (en) * | 2018-09-13 | 2019-02-19 | 西南交通大学 | A kind of detection method of high iron catenary equipotential line defective mode |
| CN109685068A (en) * | 2018-12-27 | 2019-04-26 | 上海七牛信息技术有限公司 | A kind of image processing method and system based on generation confrontation neural network |
| CN109765462A (en) * | 2019-03-05 | 2019-05-17 | 国家电网有限公司 | Fault detection method, device and the terminal device of transmission line of electricity |
-
2019
- 2019-05-28 CN CN201910451544.2A patent/CN110175571A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106339722A (en) * | 2016-08-25 | 2017-01-18 | 国网浙江省电力公司杭州供电公司 | Line knife switch state monitoring method and device |
| CN108154072A (en) * | 2016-12-02 | 2018-06-12 | 天津工业大学 | Insulator breakdown of taking photo by plane based on depth convolutional neural networks detects automatically |
| CN107220929A (en) * | 2017-06-23 | 2017-09-29 | 深圳市唯特视科技有限公司 | A kind of non-paired image method for transformation using the consistent confrontation network of circulation |
| CN108171117A (en) * | 2017-12-05 | 2018-06-15 | 南京南瑞信息通信科技有限公司 | Electric power artificial intelligence visual analysis system based on multinuclear heterogeneous Computing |
| CN108389137A (en) * | 2018-02-06 | 2018-08-10 | 国网山西省电力公司电力科学研究院 | Power fault detection early warning system based on infared spectrum technology |
| CN108665058A (en) * | 2018-04-11 | 2018-10-16 | 徐州工程学院 | A kind of generation confrontation network method based on segmentation loss |
| CN108764298A (en) * | 2018-04-29 | 2018-11-06 | 天津大学 | Electric power image-context based on single classifier influences recognition methods |
| CN109117863A (en) * | 2018-07-04 | 2019-01-01 | 全球能源互联网研究院有限公司 | The insulation subsample extending method and device of confrontation network are generated based on depth convolution |
| CN108828409A (en) * | 2018-08-03 | 2018-11-16 | 南方电网科学研究院有限责任公司 | Fault detection system based on edge calculation |
| CN109360188A (en) * | 2018-09-13 | 2019-02-19 | 西南交通大学 | A kind of detection method of high iron catenary equipotential line defective mode |
| CN109685068A (en) * | 2018-12-27 | 2019-04-26 | 上海七牛信息技术有限公司 | A kind of image processing method and system based on generation confrontation neural network |
| CN109765462A (en) * | 2019-03-05 | 2019-05-17 | 国家电网有限公司 | Fault detection method, device and the terminal device of transmission line of electricity |
Non-Patent Citations (2)
| Title |
|---|
| 杨帆: "《精通图像处理经典算法（MATLAB版）》", 28 February 2018, 北京航空航天大学出版社 * |
| 郑斌: "玻璃缺陷图像识别的关键技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027565A (en) * | 2019-12-05 | 2020-04-17 | 国网电力科学研究院武汉南瑞有限责任公司 | Power transmission and transformation equipment report content generation method based on optical vision semantic extraction |
| CN110942805A (en) * | 2019-12-11 | 2020-03-31 | 云南大学 | Insulator element prediction system based on semi-supervised deep learning |
| CN111126417A (en) * | 2020-01-17 | 2020-05-08 | 苏州浪潮智能科技有限公司 | Data center machine room management method and device |
| CN111446724A (en) * | 2020-03-17 | 2020-07-24 | 天津大学 | Multi-source interconnected micro-grid frequency coordination control method based on sliding mode algorithm |
| CN111600386A (en) * | 2020-05-18 | 2020-08-28 | 山东电工电气集团有限公司 | GIS equipment knife switch on-off state confirmation-oriented real-time monitoring system and method |
| CN112396038A (en) * | 2020-07-30 | 2021-02-23 | 国网四川省电力公司信息通信公司 | Transformer substation abnormity intelligent identification method and system based on deep learning |
| CN112396038B (en) * | 2020-07-30 | 2023-08-29 | 国网四川省电力公司信息通信公司 | Intelligent substation abnormality identification method and system based on deep learning |
| CN112836315A (en) * | 2021-02-24 | 2021-05-25 | 上海交通大学 | Neural network-based limit switch production line abnormity monitoring method |
| CN112836315B (en) * | 2021-02-24 | 2022-11-29 | 上海交通大学 | Neural network-based limit switch production line abnormity monitoring method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110175571A (en) | The intellectual monitoring of substation equipment state and recognition methods | |
| Jiang et al. | Insulator fault detection in aerial images based on ensemble learning with multi-level perception | |
| Miao et al. | Insulator detection in aerial images for transmission line inspection using single shot multibox detector | |
| CN109359697A (en) | Graph image recognition methods and inspection system used in a kind of power equipment inspection | |
| CN112199993A (en) | Method for identifying transformer substation insulator infrared image detection model in any direction based on artificial intelligence | |
| CN102143354B (en) | Method for recognizing and calculating galloping of transmission conductor based on video image processing | |
| CN101833673A (en) | Electric power switchgear switch state image recognition system | |
| CN113486877B (en) | Power equipment infrared image real-time detection and diagnosis method based on light artificial intelligent model | |
| CN112651516A (en) | Intelligent inspection system and method for transformer substation machine room based on AR technology | |
| CN116824517B (en) | Substation operation and maintenance safety control system based on visualization | |
| CN112733914A (en) | Underwater target visual identification and classification method based on support vector machine | |
| CN116486240A (en) | Application of image recognition algorithm in intelligent inspection method of unmanned aerial vehicle of power transmission line | |
| CN116245871A (en) | Power transmission line abnormal target detection method based on improved YOLOX algorithm | |
| CN113496210B (en) | Photovoltaic string tracking and fault tracking method based on attention mechanism | |
| CN109509169A (en) | The pressing plate condition checkout gear and method of learning distance metric based on nearest neighbour classification | |
| CN106355187A (en) | Application of visual information to electrical equipment monitoring | |
| Daogang et al. | Anomaly identification of critical power plant facilities based on YOLOX-CBAM | |
| Xia et al. | A switch state recognition method based on improved VGG19 network | |
| CN104166831A (en) | ALBP and SRC algorithm-based fatigue detection method and system | |
| Dong et al. | An Automatic Detection Method of Bird's Nest on Electric Tower Based on Attention Full Convolutional Neural Networks | |
| CN114355110A (en) | Fault current mode identification method based on convolutional neural network | |
| CN114821042A (en) | R-FCN disconnecting link detection method combining local features and global features | |
| CN113139476A (en) | Data center-oriented human behavior attribute real-time detection method and system | |
| Shan et al. | Research on Efficient Detection Method of Foreign Objects on Transmission Lines Based on Improved YOLOv4 Network | |
| Fu et al. | Study of a Detection and Recognition Algorithm for High-Voltage Switch Cabinet Based on Deep Learning with an Improved Faster-RCNN |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 413002 No. 1 Xuefu Road West, Longling Industrial Park, Heshan District, Yiyang City, Hunan Province Applicant after: Huaxiang XiangNeng Technology Co.,Ltd. Applicant after: HUNAN University Address before: 413002 No. 1 Xuefu Road West, Longling Industrial Park, Heshan District, Yiyang City, Hunan Province Applicant before: HXXN ELECTRICITY Co.,Ltd. Applicant before: HUNAN University |
|
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210219 Address after: 413002 No. 1 Xuefu Road West, Longling Industrial Park, Heshan District, Yiyang City, Hunan Province Applicant after: Huaxiang XiangNeng Technology Co.,Ltd. Address before: 413002 No. 1 Xuefu Road West, Longling Industrial Park, Heshan District, Yiyang City, Hunan Province Applicant before: Huaxiang XiangNeng Technology Co.,Ltd. Applicant before: HUNAN University |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190827 |