CN115731455A - Looped network metering cabinet - Google Patents
Looped network metering cabinet Download PDFInfo
- Publication number
- CN115731455A CN115731455A CN202211337909.7A CN202211337909A CN115731455A CN 115731455 A CN115731455 A CN 115731455A CN 202211337909 A CN202211337909 A CN 202211337909A CN 115731455 A CN115731455 A CN 115731455A
- Authority
- CN
- China
- Prior art keywords
- feature map
- detection
- convolution
- training
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Remote Monitoring And Control Of Power-Distribution Networks (AREA)
Abstract
The application discloses looped network measurement cabinet. Firstly, passing an acquired plugging state detection image and a reference plugging state image through a twin network model comprising a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map, then respectively passing the detection feature map and the reference feature map through a parallel weight distribution module to obtain a detection enhancement feature map and a reference enhancement feature map, then calculating a difference feature map between the detection enhancement feature map and the reference enhancement feature map, and finally passing the difference feature map through a classifier to obtain a classification result for indicating whether the plugging state of the metering current transformer meets a preset requirement. Through the mode, the plugging state of the current transformer can be effectively monitored in real time, so that faults and accidents are avoided, and the normal and safe work of the ring network metering cabinet is ensured.
Description
Technical Field
The application relates to the technical field of intelligent monitoring, and more specifically relates to a looped network metering cabinet capable of effectively monitoring the plugging state of a current transformer in real time.
Background
With the continuous promotion of lean management, line loss is increasingly regarded as an important assessment index. In recent years, the management targets of refining low-voltage line loss management and control and promoting 10kV branching are provided, the examination gateways such as the healthy 10kV line contact points are required, and the gateways are provided with bidirectional metering devices to realize full coverage and full collection of gateway metering.
At present, the 'cable entering the ground' of an urban power grid becomes the mainstream, the outdoor ring main unit is widely applied, and a switch is located in the outdoor ring main unit. The fact that bidirectional metering is achieved through the power grid 10kV switch means that a metering device needs to be added to the ring main unit. However, if the metering points are added to the ring main unit according to the traditional high-voltage metering scheme, the problems that the space of the ring main unit is narrow and the high-voltage transformer cannot be installed can be met.
Therefore, an optimized ring network metering cabinet is expected.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a looped network measurement cabinet. Firstly, passing an acquired plugging state detection image and a reference plugging state image through a twin network model comprising a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map, then respectively passing the detection feature map and the reference feature map through a parallel weight distribution module to obtain a detection enhancement feature map and a reference enhancement feature map, then calculating a difference feature map between the detection enhancement feature map and the reference enhancement feature map, and finally passing the difference feature map through a classifier to obtain a classification result for indicating whether the plugging state of the metering current transformer meets a preset requirement. Through the mode, the plugging state of the current transformer can be effectively monitored in real time, so that faults and accidents are avoided, and the normal and safe work of the ring network metering cabinet is ensured.
According to an aspect of the present application, there is provided a looped network metering cabinet, comprising:
the plug state monitoring unit is used for acquiring a plug state detection image and a reference plug state image of the metering current transformer;
the plug state encoding unit is used for enabling the plug state detection image and the reference plug state image to pass through a twin network model comprising a first convolutional neural network and a second convolutional neural network so as to obtain a detection characteristic diagram and a reference characteristic diagram, wherein the first convolutional neural network and the second convolutional neural network have the same network structure;
the feature enhancement unit is used for enabling the detection feature map and the reference feature map to pass through a parallel weight distribution module respectively so as to obtain a detection enhancement feature map and a reference enhancement feature map;
a difference characterization unit, configured to calculate a difference feature map between the detected enhanced feature map and the reference enhanced feature map; and the monitoring result generating unit is used for enabling the differential characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the plugging state of the metering current transformer meets the preset requirement or not.
Compared with the prior art, according to the looped network metering cabinet, firstly, the obtained plugging state detection image and the reference plugging state image pass through a twin network model comprising a first convolution neural network and a second convolution neural network to obtain a detection characteristic diagram and a reference characteristic diagram, then the detection characteristic diagram and the reference characteristic diagram pass through a parallel weight distribution module respectively to obtain a detection enhancement characteristic diagram and a reference enhancement characteristic diagram, then, a difference characteristic diagram between the detection enhancement characteristic diagram and the reference enhancement characteristic diagram is calculated, and finally, the difference characteristic diagram passes through a classifier to obtain a classification result for indicating whether the plugging state of the metering current transformer meets the preset requirement or not. Through such a mode, can carry out real time monitoring effectively to current transformer's plug state, and then avoid the emergence of trouble and accident, guarantee the normal safe work of looped netowrk metering cabinet.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is an application scenario diagram of a ring network metering cabinet according to an embodiment of the present application.
Fig. 2 is a schematic block diagram of a ring network metering cabinet according to an embodiment of the present application.
Fig. 3 is a schematic block diagram of the plug-in/pull-out status encoding unit in the ring network metering cabinet according to the embodiment of the present application.
Fig. 4 is a schematic block diagram of the feature enhancing unit in the ring network metering cabinet according to the embodiment of the present application.
Fig. 5 is a schematic block diagram of a training module further included in the ring network metering cabinet according to the embodiment of the present application.
Fig. 6 is a flowchart of a method for monitoring a plugging/unplugging state of a metering current transformer according to an embodiment of the present application.
Fig. 7 is a schematic diagram of a system architecture of a method for monitoring a plugging/unplugging state of a metering current transformer according to an embodiment of the present disclosure.
Fig. 8 is a front view of a totally-enclosed and totally-insulated 10kV looped network type metering box according to an embodiment of the application.
Fig. 9 is a left side view of a fully sealed and fully insulated 10kV looped network type metering box according to an embodiment of the present application.
Fig. 10 is a sectional view of the power PT cabinet of fig. 8.
Fig. 11 is a left side view of the incoming line breaker cabinet of fig. 8.
Fig. 12 is a cross-sectional view of the incoming circuit breaker cabinet of fig. 8.
Figure 13 is a left side view of the outlet circuit breaker cabinet of figure 8.
Figure 14 is a cross-sectional view of the outgoing line breaker cabinet of figure 8.
Fig. 15 is a schematic back view of the ring network metering cabinet.
In the figure: 1. the system comprises a secondary room indication panel, 2, an operation mechanism panel, 3, a lower cabinet door, 4, a voltmeter, 5, an ammeter, 6, an intelligent protection controller, 7, a grounding bar, 8, a lower cabinet, 9, a mechanism cabin, 10, a secondary room, 11, a battery box, 12, an air box, 13, a three-position load switch operation mechanism, 14, a three-position load switch, 15, a three-phase integrated PT,16, a PT cable, 17, a metering PT room, 18, a grounding wire, 19, a parallel cabinet inner cone sleeve, 20, a lifting sling, 21, a plug-in metering PT,22, an inner cone sleeve, 23, a circuit breaker, 24, a circuit breaker operation mechanism, 25, an isolating switch operation mechanism, 26, a three-position isolating switch, 27, a wire inlet sleeve, 28, a lightning protection cable, 29, a metering CT room, 30, an instrument box, 31, a plug-in metering CT,32, an inner cone sleeve, 33, a meter, 34, a wire outlet sleeve, 35, a wire outlet, a wire, a three-phase cable, 37 and a three-phase integrated mutual inductor.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Overview of scenes
As mentioned above, at present, the urban power grid "cable is in the ground" is the mainstream, the outdoor ring main unit is widely used, and the switch is located in the outdoor ring main unit. The fact that bidirectional metering is achieved through the power grid 10kV switch means that a metering device needs to be added to the ring main unit. However, if the metering points are added to the ring main unit according to the traditional high-voltage metering scheme, the problems that the space of the ring main unit is narrow and the high-voltage transformer cannot be installed can be met.
In order to solve the technical problem, the applicant of the present application tries to install a metering current transformer and an electric energy meter compartment at the back of each outlet switch of the looped network metering cabinet, and each outlet is separately metered without mutual influence. Moreover, the metering current transformer adopts a plug-in structure, although the installation mode is convenient for field validation and capacity increase operation, in the actual operation process, the plug-in structure is found to loosen along with the increase of the service time (for example, due to continuous slight vibration), so that the fault is caused. Therefore, monitoring of the plugging and unplugging state of the metering current transformer is expected, and when the plugging and unplugging state is judged to be abnormal, an early warning prompt is generated, so that faults are avoided, and the effect of fault prevention is achieved.
Specifically, in the technical scheme of the application, an artificial intelligence technology based on deep learning is adopted to extract a characteristic difference between a plugging state detection image and an image (reference plugging state image) with a good plugging state of the metering current transformer, so as to judge whether the plugging state of the metering current transformer meets a preset requirement. Namely, the plugging and unplugging state of the current transformer is effectively monitored in real time by using a visual monitoring algorithm in an artificial intelligence technology, so that an early warning prompt is generated when the plugging and unplugging state of the current transformer is monitored to avoid the occurrence of faults and accidents, and the normal and safe operation of the ring network metering cabinet is ensured.
Specifically, in the technical scheme of the application, firstly, a plug-in state detection image and a reference plug-in state image of the metering current transformer are obtained through a camera. Then, a convolutional neural network model with excellent performance in terms of implicit feature extraction of images is used to extract features of the plug state detection image and the reference plug state image, however, when the plug state detection image of the metering current transformer is actually acquired, it is considered that a difficulty is brought to subsequent feature differences due to a visual angle difference which may exist during shooting. Therefore, in order to make the convolutional neural network model insensitive to the deviation of the view angle, the convolution kernels of the layers of the convolutional neural network model are designed as asymmetric convolution kernels.
Specifically, the plug state detection image and the reference plug state image are passed through a twin network model including a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map, where the first convolutional neural network and the second convolutional neural network have the same network structure, and both the first convolutional neural network and the second convolutional neural network employ asymmetric convolution kernels to process the plug state detection image and the reference plug state image respectively. It should be understood that the asymmetric convolution kernel is obtained by performing convolution operation on the same feature through three different convolution kernels, wherein the length and width of the convolution kernels are H × W, 1 × W and H × 1. The entry point of the method is to obtain better feature expression through an asymmetric convolution kernel, so that the richness of feature map information can be improved, and the robustness of the model to the bar code target overturning and rotating is improved.
Further, it should be understood that, when monitoring the plugging/unplugging state of the metering current transformer, the hidden feature information at the spatial position and the channel dimension of the plugging/unplugging position of the metering current transformer should be focused on, and useless interference features irrelevant to the plugging/unplugging state detection of the metering current transformer should be omitted. Therefore, in the technical solution of the present application, for the problem that the target detection accuracy is low due to edge blurring in the detection feature map and the reference feature map, a parallel weight assignment module is used to perform feature enhancement in the detection feature map and the reference feature map. Specifically, the detection feature map and the reference feature map are respectively passed through a parallel weight distribution module to obtain a detection enhanced feature map and a reference enhanced feature map, so that effective feature representation can be enhanced, useless feature information can be suppressed, and accuracy of subsequent classification can be improved. In particular, here, the parallel weight assignment module performs feature enhancement on the detected feature map and the reference feature map by using a spatial attention module and a channel attention module, respectively, wherein the extracted image features reflect the correlation and importance among feature channels, and the extracted image features reflect the weight of spatial dimension feature differences to suppress or enhance features at different spatial positions.
Then, calculating a difference feature map between the detection enhancement feature map and the reference enhancement feature map to express a feature difference between a plugging state detection image and a plugging state good image (reference plugging state image) of the metering current transformer, and further performing classification judgment on the plugging state of the metering current transformer according to the feature difference. Namely, the differential characteristic diagram is classified in a classifier to obtain a classification result for indicating whether the plugging state of the metering current transformer meets the preset requirement. Therefore, whether the plugging and unplugging state of the metering current transformer meets the preset requirement or not can be accurately judged.
Particularly, in the technical solution of the present application, since the differential feature map as the classification feature map is obtained by calculating the difference between the detection enhancement feature map and the reference enhancement feature map, in the training process, the classification loss function of the classifier respectively passes through the first convolutional neural network and the second convolutional neural network when the gradient is reversely propagated, so that the first convolutional neural network may resolve the image semantics of the plug-state detection image and the feature extraction pattern of the image semantics of the reference plug-state image due to abnormal gradient divergence, thereby affecting the accuracy of the classification result of the classification feature map.
Therefore, preferably, a suppression loss function for feature extraction pattern resolution of the detection enhancement feature map and the reference enhancement feature map is introduced, expressed as:
here, V 1 And V 2 Respectively, the feature vectors obtained after the detection enhanced feature map and the reference enhanced feature map are expanded, M 1 And M 2 Respectively, the classifier is for the feature vector V 1 And V 2 Is given by the weight matrix, | · | F Represents the F norm of the matrix, an
Representing the square of the two norms of the vector.
Specifically, the suppression loss function of feature extraction pattern resolution ensures that the directional derivative in the reverse gradient propagation is normalized near the branch point of the gradient propagation by making the difference distribution of the classifier in the form of cross entropy in relation to the weight matrix of the feature vector obtained after the expansion of the detection enhanced feature map and the reference enhanced feature map be consistent with the true feature difference distribution of the feature vector, that is, the gradient is weighted for the feature extraction patterns of the first convolutional neural network and the second convolutional neural network. Therefore, the resolution of the feature extraction mode is inhibited, the feature extraction capability of the first convolutional neural network on the image semantics of the plug state detection image and the feature extraction capability of the second convolutional neural network on the image semantics of the reference plug state image are improved, and the accuracy of the classification result of the classification feature map is correspondingly improved. Therefore, the plugging state of the current transformer can be effectively monitored in real time, an early warning prompt is generated when the plugging state of the current transformer is monitored to be abnormal, and then faults and accidents are avoided, so that the normal and safe operation of the ring network metering cabinet is ensured.
Based on this, this application provides a looped network measurement cabinet, and it includes: the plug state monitoring unit is used for acquiring a plug state detection image and a reference plug state image of the metering current transformer; the plug state encoding unit is used for enabling the plug state detection image and the reference plug state image to pass through a twin network model comprising a first convolutional neural network and a second convolutional neural network so as to obtain a detection characteristic diagram and a reference characteristic diagram, wherein the first convolutional neural network and the second convolutional neural network have the same network structure; the feature enhancement unit is used for enabling the detection feature map and the reference feature map to pass through a parallel weight distribution module respectively to obtain a detection enhancement feature map and a reference enhancement feature map; a difference characterization unit, configured to calculate a difference feature map between the detected enhanced feature map and the reference enhanced feature map; and the monitoring result generating unit is used for enabling the differential characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the plugging state of the metering current transformer meets the preset requirement or not.
In the application, the measurement current transformer and the electric energy meter compartment are installed on the back of each path of outlet switch of the ring network measurement cabinet by an applicant, each path of outlet is separately measured, the measurement current transformer and the measurement PT sensor are of a plug-in structure, and the plug-in state of the measurement current transformer or the measurement PT sensor is further monitored, so that when the plug-in state is judged to be abnormal, an early warning prompt is generated, the occurrence of faults caused by looseness of the plug-in structure along with the increase of the service time is avoided, and the effect of fault prevention is achieved.
It is worth mentioning that, the looped netowrk measurement cabinet described in this application can be the outdoor high-voltage ring net type electric-energy metering box of RMKR-12 type, and the outdoor high-voltage ring net type electric-energy metering box of RMKR-12 type is a combined type metering device that satisfies national network customization looped netowrk cabinet dimensional requirement and accords with a secondary and fuses the requirement. It has the following characteristics: has the following characteristics: 1) The vacuum break full-sealed full-insulation structure is adopted, and the high-voltage electrified body is completely sealed in SF6 gas, so that the safety of personnel is protected to the maximum extent; 2) A metering current transformer and an electric energy meter compartment are arranged at the back of each path of outgoing line switch, and each path of outgoing line is independently metered without mutual influence; 3) The metering current transformer and the metering PT adopt a plug-in structure, so that the requirements of full sealing and full insulation are met, the operation of field effect test and capacity increase is facilitated, and the failure rate is greatly reduced; 4) The transformation ratio of the metering current transformer can meet the full-range coverage of 5/5A to 600/5A, and can be flexibly selected by a user; 5) The multi-input multi-output scheme can be realized for the user to select flexibly. Typical schemes are 2 in and 2 out, 2 in and 4 out, 2 in and 6 out, and the like; 6) Subsequent upgrades can be implemented. The DTU position is reserved on the right side, and subsequent automatic upgrading can be realized; 7) The metering PT is connected out from the bus, and the PT is in a switch-on state as long as the bus is electrified, so that the phenomenon of less electric energy caused by misoperation is avoided.
Further, referring to fig. 8 to 14, a looped network metering cabinet exemplified in the present application will be described. As shown in fig. 8 to 14, a totally enclosed and totally insulated 10kV looped network metering box comprises a power supply PT cabinet, an incoming line breaker cabinet, an outgoing line breaker cabinet and a DTU cabinet. The cabinet body, power PT cabinet, inlet wire circuit breaker cabinet and the circuit breaker cabinet of being qualified for the next round of competitions pass through and are connected by taper sleeve 19 in the cabinet. The cabinet body, power PT cabinet, inlet wire circuit breaker cabinet and the circuit breaker cabinet that is qualified for the next round of competitions are inside to be separated into independent compartment through metal partition, and the front is operation interface, and the back is metering interface. The cabinet body, inside three-phase integral type PT15 and the PT cable 16 of being provided with of power PT cabinet, inside inlet wire sleeve pipe 27, three-phase integral type mutual-inductor 37, arrester 35 and the incoming cable 28 of being provided with of inlet wire circuit breaker cabinet, the inside outlet wire sleeve pipe 34, three-phase integral type mutual-inductor 37, arrester 35 and the outlet cable 36 of being provided with of outlet wire circuit breaker cabinet, the inside plug-in measurement PT that is provided with of inlet wire circuit breaker, the inside plug-in measurement CT that is provided with of outlet wire circuit breaker.
In this embodiment, the whole case of 10kV looped network batch meter adopts stainless steel material, and the cabinet is divided into different independent compartments through the metal partition plate, when a fault occurs, the fault does not reach the adjacent compartments, the front side is an operation interface, and the back side is a metering interface. The responsibility range of everybody is separated from the design structure, the mutual interference and the skin tearing are reduced, the metering CT and the metering PT are arranged below a metering interface and both adopt plug-in type structures, the plug-in structural formula of the PT is a mature design structure, and the plug-in structural formula of the CT is a brand new design, so that the insulation reliability is ensured, and the field replacement is convenient; the metering box and the metering equipment are both provided with padlocks and lead sealing structures and are both provided with observation windows. CT and PT room still have the electromagnetic lock, guarantee under the electrified circumstances of equipment, the cabinet door can not opened, has guaranteed personal safety. The secondary chamber is provided with intelligent protection controller, possesses perfect syllogic protect function. Be equipped with the light in the strapping table case, be provided with the button on the panel, need not open lead sealing and just can realize the on-the-spot meter reading, greatly reduced meter reading personnel's intensity of labour, improved meter reading efficiency.
Further, referring to fig. 8 to fig. 14, another ring network metering cabinet exemplified in the present application is described. As shown in fig. 8 to 14, a fully-sealed and fully-insulated 10kV looped network metering box comprises a power supply PT cabinet, an incoming line breaker cabinet, an outgoing line breaker cabinet and a DTU cabinet, wherein the power supply PT cabinet comprises a lower cabinet body 8, a mechanism bin 9, a secondary chamber 10, a battery box 12 and an air box 11, wherein the lower cabinet body is arranged at the lower part of the power supply PT cabinet, a three-phase integrated PT15 and a PT cable 16 are arranged inside the lower cabinet body 8, and a grounding bar 7 is arranged at the lower part of the lower cabinet body 8; the mechanism bin 9 is arranged at the upper part of the lower cabinet body 8, and a three-station load switch operating mechanism is arranged in the mechanism bin 9; the inside of the secondary chamber 10 is provided with a voltmeter 4, the battery box 12 is arranged at the upper part of the air box 11, the inside of the air box 11 is provided with a three-position load switch 14, and the upper part of the air box 11 is provided with a combined cabinet inner cone sleeve 19 and a lifting rib 20. The incoming line breaker cabinet comprises a lower cabinet body 8, a mechanism bin 9, a secondary chamber 10, a metering PT chamber 17 and an air box 11, wherein the lower cabinet body 8 is arranged at the lower part of the incoming line breaker cabinet, a three-phase integrated mutual inductor 37, an incoming line sleeve 27, a lightning arrester 35 and an incoming line cable 28 are arranged inside the lower cabinet body 8, and a grounding bar 7 is arranged at the lower part of the lower cabinet body 8; the mechanism bin 9 is arranged at the upper part of the lower cabinet body 8, and a three-station load switch operating mechanism and a circuit breaker operating mechanism 24 are arranged in the mechanism bin 9; an ammeter 5 and an intelligent protection controller 6 are arranged in the secondary chamber 10; the measuring PT chamber 17 is arranged at the upper part of the air box 11, and the measuring PT chamber 17 is provided with a plug-in type measuring PT and a grounding wire 18; a three-position isolating switch 26 and a circuit breaker 23 are arranged in the gas tank 11, and a combined inner cone sleeve 19 and a lifting rib 20 are arranged at the upper part of the gas tank 11. The outgoing line breaker cabinet comprises a lower cabinet body 8, a mechanism bin 9, a secondary chamber 10, a metering CT chamber 29 and an air box 11, wherein the lower cabinet body 8 is arranged at the lower part of the outgoing line breaker cabinet, a three-phase integrated mutual inductor 37, an outgoing line sleeve 34, a lightning arrester 35 and an outgoing line cable 34 are arranged inside the lower cabinet body 8, and a grounding bar 7 is arranged at the lower part of the lower cabinet body 8; the mechanism bin 9 is arranged at the upper part of the lower cabinet body 8, and a three-station load switch operating mechanism and a circuit breaker operating mechanism 24 are arranged in the mechanism bin 9; an ammeter 5 and an intelligent protection controller 6 are arranged in the secondary chamber 10; the measurement CT room 29 is arranged at the upper part of the air box 11, and the measurement CT room 29 is provided with a plug-in measurement CT and a grounding wire 18; a three-position isolating switch 26 and a circuit breaker 23 are arranged in the gas tank 11, and a combined inner cone sleeve 19 and a lifting rib 20 are arranged at the upper part of the gas tank 11.
In this embodiment, 10kV looped netowrk batch meter possesses looped netowrk cabinet and batch meter function simultaneously, has that the scheme is nimble, independent assortment and multichannel measures the characteristics respectively.
In the embodiment, the protection device on the front side of the 10kV ring network metering box also has a unique anti-electricity-stealing function; through the current secondary collection to measurement current transformer, then compare often with protection current transformer's electric current, when finding that the load has the anomaly, can send the warning automatically, the information of warning is sent on measurement personnel's the cell-phone and on the control room computer through wireless communication. The measurement personnel can in time inspect the batch meter according to information to avoid the economic loss that long-term electric larceny action caused the power company.
Further, fig. 1 illustrates an application scenario diagram of the ring network metering cabinet according to an embodiment of the present application. As shown in fig. 1, in the application scenario, a plugging state detection image (e.g., D1 as illustrated in fig. 1) and a reference plugging state image (e.g., D2 as illustrated in fig. 1) of a metering current transformer (e.g., F as illustrated in fig. 1) are obtained through a camera (e.g., C as illustrated in fig. 1), and then the obtained plugging state detection image and the obtained reference plugging state image are input into a server (e.g., S as illustrated in fig. 1) in which a metering current transformer plugging state monitoring algorithm is deployed, where the server is capable of processing the plugging state detection image and the reference plugging state image by using the metering current transformer plugging state monitoring algorithm to generate a classification result for indicating whether the plugging state of the metering current transformer meets a predetermined requirement.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary Ring network metering cabinet
Fig. 2 illustrates a block diagram schematic diagram of a ring network metering cabinet according to an embodiment of the present application. As shown in fig. 2, the ring network metering cabinet 100 according to the embodiment of the present application includes: the plugging state monitoring unit 110 is configured to obtain a plugging state detection image and a reference plugging state image of the metering current transformer; a plug state encoding unit 120, configured to pass the plug state detection image and the reference plug state image through a twin network model including a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map, where the first convolutional neural network and the second convolutional neural network have the same network structure; a feature enhancing unit 130, configured to pass the detected feature map and the reference feature map through a parallel weight assignment module respectively to obtain a detected enhanced feature map and a reference enhanced feature map; a difference characterization unit 140, configured to calculate a difference feature map between the detected enhanced feature map and the reference enhanced feature map; and a monitoring result generating unit 150, configured to pass the differential feature map through a classifier to obtain a classification result, where the classification result is used to indicate whether the plugging/unplugging state of the metering current transformer meets a predetermined requirement.
More specifically, in the embodiment of the present application, the plugging status monitoring unit 110 is configured to obtain a plugging status detection image and a reference plugging status image of the metering current transformer. Although the metering current transformer adopts a plug-in structure to facilitate field validation and capacity increase operations, in the actual operation process, the plug-in structure can loosen along with the increase of the service time (for example, due to continuous slight vibration), and further faults are caused. Therefore, the plugging state detection image and the reference plugging state image (image with good plugging state) of the metering current transformer are obtained, the plugging state of the metering current transformer is monitored, whether the plugging state of the metering current transformer meets the preset requirement or not is judged, and an early warning prompt is generated when the plugging state is judged to be abnormal.
More specifically, in this embodiment of the application, the plug state encoding unit 120 is configured to pass the plug state detection image and the reference plug state image through a twin network model including a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map, where the first convolutional neural network and the second convolutional neural network have the same network structure. The plug-in state detection image and the reference plug-in state image are extracted by using a convolutional neural network model with excellent performance in the aspect of implicit feature extraction of images, however, when the plug-in state detection image of the metering current transformer is actually collected, the subsequent feature difference is difficult due to the visual angle difference possibly existing during shooting. Therefore, in order to make the convolutional neural network model insensitive to the deviation of the view angle, the convolution kernels of the layers of the convolutional neural network model are designed as asymmetric convolution kernels. The first convolutional neural network and the second convolutional neural network respectively process the plugging state detection image and the reference plugging state image by adopting asymmetric convolutional kernels. It should be understood that the asymmetric convolution kernel is obtained by performing convolution operation on the same feature through three different convolution kernels, wherein the length and width of the convolution kernels are H × W, 1 × W and H × 1. The entry point of the method is to obtain better feature expression through an asymmetric convolution kernel, so that the richness of feature map information can be improved, and the robustness of the model to the bar code target overturning and rotating is improved.
Accordingly, as shown in fig. 3, in a specific example, the plugging status encoding unit 120 includes: a detection image coding subunit 121, configured to perform, in a layer forward pass, the following operations on the input data respectively using the layers of the first convolutional neural network in the twin network model: performing convolution processing on the input data based on a two-dimensional convolution kernel to obtain a first convolution feature map; performing one-dimensional convolution processing on the input data based on a first one-dimensional convolution kernel to obtain a second convolution characteristic diagram; performing one-dimensional convolution processing on the input data based on a second one-dimensional convolution kernel to obtain a third convolution characteristic diagram; fusing the first convolution feature map, the second convolution feature map and the third convolution feature map to obtain a fused convolution feature map; pooling the fused convolution feature map to obtain a pooled feature map; carrying out nonlinear activation processing on the pooled feature map to obtain an activated feature map; the output of the last layer of the first convolutional neural network in the twin network model is the detection feature map, and the input of the first layer of the first convolutional neural network in the twin network model is the plugging and unplugging state detection image.
Accordingly, as shown in fig. 3, in a specific example, the plugging status encoding unit 120 includes: a reference image encoding subunit 122, configured to perform, in a layer forward pass, the following on the input data respectively using the layers of the second convolutional neural network in the twin network model: performing convolution processing on the input data based on a two-dimensional convolution kernel to obtain a first convolution feature map; performing one-dimensional convolution processing on the input data based on a first one-dimensional convolution kernel to obtain a second convolution characteristic diagram; performing one-dimensional convolution processing on the input data based on a second one-dimensional convolution kernel to obtain a third convolution characteristic diagram; fusing the first convolution feature map, the second convolution feature map and the third convolution feature map to obtain a fused convolution feature map; pooling the fused convolution feature map to obtain a pooled feature map; carrying out nonlinear activation processing on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network in the twin network model is the reference characteristic diagram, and the input of the first layer of the second convolutional neural network in the twin network model is the reference plugging state image.
More specifically, in the embodiment of the present application, the feature enhancing unit 130 is configured to pass the detected feature map and the reference feature map through a parallel weight assignment module to obtain a detected enhanced feature map and a reference enhanced feature map, respectively. Therefore, effective feature representation can be enhanced, useless feature information can be restrained, and accuracy of subsequent classification can be improved. It should be understood that, when monitoring the plugging and unplugging state of the metering current transformer, the hidden feature information at the spatial position and the channel dimension of the plugging and unplugging position of the metering current transformer should be focused on, and useless interference features irrelevant to the plugging and unplugging state detection of the metering current transformer are omitted. Therefore, in the technical solution of the present application, for the problem that the target detection accuracy is low due to edge blurring in the detection feature map and the reference feature map, a parallel weight assignment module is used to perform feature enhancement in the detection feature map and the reference feature map. In particular, here, the parallel weight assignment module performs feature enhancement on the detected feature map and the reference feature map by using a spatial attention module and a channel attention module, respectively, wherein the extracted image features reflect the correlation and importance among feature channels, and the extracted image features reflect the weight of spatial dimension feature differences to suppress or enhance features at different spatial positions.
Accordingly, as shown in fig. 4, in a specific example, the feature enhancing unit 130 includes: a spatial attention branching subunit 131, configured to pass the detection feature map and the reference feature map through a spatial attention module of a parallel weight assignment module respectively to obtain a detection spatial attention feature map and a reference spatial attention feature map; a channel attention branching subunit 132, configured to pass the detection feature map and the reference feature map through a channel attention module of the parallel weight assignment module respectively to obtain a detection channel attention feature map and a reference channel attention feature map; and a fusion subunit 133, configured to fuse the detection spatial attention feature map and the detection channel attention feature map to obtain the detection enhancement feature map, and fuse the reference spatial attention feature map and the reference channel attention feature map to obtain the reference enhancement feature map.
Accordingly, in a specific example, the spatial attention branching subunit 131 is further configured to: depth convolution coding is carried out on the detection feature map and the reference feature map respectively by using a convolution coding part of a space attention module of the parallel weight distribution module so as to obtain a detection convolution feature map and a reference convolution feature map; inputting the detection convolution feature map and the reference convolution feature map into a spatial attention portion of a spatial attention module of the parallel weight assignment module respectively to obtain a detection spatial attention map and a reference spatial attention map; respectively passing the detection space attention diagram and the reference space attention diagram through a Softmax activation function to obtain a detection space attention feature map and a reference space attention feature map; and calculating the point-by-point multiplication of the detection space attention feature map and the detection feature map to obtain the detection space attention feature map, and calculating the point-by-point multiplication of the reference space attention feature map and the reference feature map to obtain the reference space attention feature map.
Accordingly, in one specific example, the channel attention branching subunit 132 is further configured to: inputting the detection feature map and the reference feature map into the multilayer convolution layer of the channel attention module of the parallel weight distribution module respectively to obtain a detection convolution feature map and a reference convolution feature map; respectively calculating the global mean value of each feature matrix of the detection convolution feature map and the reference convolution feature map along the channel dimension to obtain a detection channel feature vector and a reference channel feature vector; inputting the detection channel feature vector and the reference channel feature vector into the Sigmoid activation function respectively to obtain a detection channel attention weight vector and a reference channel attention weight vector; and weighting each feature matrix along the channel dimension of the detection convolution feature map by taking the feature value of each position in the detection channel attention weight vector as a weight to obtain the detection channel attention feature map, and weighting each feature matrix along the channel dimension of the reference convolution feature map by taking the feature value of each position in the reference channel attention weight vector as a weight to obtain the reference channel attention feature map.
More specifically, in the embodiment of the present application, the difference characterization unit 140 is configured to calculate a difference feature map between the detected enhanced feature map and the reference enhanced feature map. And calculating a difference characteristic diagram between the detection enhanced characteristic diagram and the reference enhanced characteristic diagram to express the characteristic difference between the plugging and unplugging state detection image of the metering current transformer and an image (reference plugging and unplugging state image) with a good plugging and unplugging state, and further carrying out classification judgment on the plugging and unplugging state of the metering current transformer by using the characteristic difference.
Accordingly, in a specific example, the difference characterization unit 140 is further configured to: calculating a difference feature map between the detection enhancement feature map and the reference enhancement feature map in the following formula; wherein the formula is:
wherein, F 1 For said detection of enhanced feature maps, F 2 For the purpose of the reference enhancement feature map,
indicating a difference by position.
More specifically, in this embodiment of the application, the monitoring result generating unit 150 is configured to pass the differential feature map through a classifier to obtain a classification result, where the classification result is used to indicate whether the plugging/unplugging state of the metering current transformer meets a predetermined requirement.
Accordingly, in a specific example, the monitoring result generating unit 150 is further configured to: processing the differential feature map by using the classifier according to the following formula to obtain the classification result, wherein the formula is as follows: o = softmax { (W) n ,B n ):…:(W 1 ,B 1 )|Project(F d ) Wherein, project (F) d ) Representing the projection of the difference profile as a vector, W 1 To W n As a weight matrix for each fully connected layer, B 1 To B n A bias matrix representing the layers of the fully connected layer.
Accordingly, in a specific example, the ring network metering cabinet further includes a training module for training the twin network model, the parallel weight distribution module, and the classifier; as shown in fig. 5, the training module 200 includes: a training plug state monitoring unit 210, configured to obtain training data, where the training data includes a training plug state detection image and a training reference plug state image of the metering current transformer, and whether the plug state of the metering current transformer meets a true value of a predetermined requirement; a training plug state encoding unit 220, configured to pass the training plug state detection image and the training reference plug state image through the twin network model including the first convolutional neural network and the second convolutional neural network to obtain a training detection feature map and a training reference feature map, where the first convolutional neural network and the second convolutional neural network have the same network structure; a training feature enhancement unit 230, configured to pass the training detection feature map and the training reference feature map through the parallel weight assignment module respectively to obtain a training detection enhancement feature map and a training reference enhancement feature map; a training difference characterization unit 240, configured to calculate a training difference feature map between the training detection enhanced feature map and the training reference enhanced feature map; a classification loss unit 250, configured to pass the training difference feature map through the classifier to obtain a classification loss function value; an inhibition loss function value calculation unit 260, configured to calculate an inhibition loss function value of the feature extraction mode resolution of the training detection enhanced feature map and the training reference enhanced feature map; and a training unit 270 for training the twin network model, the parallel weight assignment module and the classifier with a weighted sum of the suppression loss function values and the classification loss function values of the feature extraction mode resolution as the loss function values.
Particularly, in the technical solution of the present application, since the differential feature map as the classification feature map is obtained by calculating the difference between the detection enhancement feature map and the reference enhancement feature map, in the training process, the classification loss function of the classifier respectively passes through the first convolutional neural network and the second convolutional neural network when the gradient is reversely propagated, so that the first convolutional neural network may resolve the image semantics of the plug-state detection image and the feature extraction pattern of the image semantics of the reference plug-state image due to abnormal gradient divergence, thereby affecting the accuracy of the classification result of the classification feature map. Therefore, it is preferable to introduce a suppression loss function for feature extraction pattern resolution of the detection enhancement feature map and the reference enhancement feature map.
Accordingly, in a specific example, the suppression loss function value calculating unit 260 is further configured to: calculating an inhibition loss function value of the feature extraction mode resolution of the training detection enhancement feature map and the training reference enhancement feature map according to the following formula; wherein the formula is:
wherein V 1 And V 2 Respectively, the feature vectors obtained after the detection enhanced feature map and the reference enhanced feature map are expanded, and M is 1 And M 2 Respectively, the classifier is used for obtaining a weight matrix of feature vectors after the detection enhanced feature map and the reference enhanced feature map are developed,
represents the square of the two-norm of the vector, | - | F The F-norm of the matrix is represented,
representing a position-wise subtraction and log represents a base-2 logarithmic function.
Specifically, the suppression loss function of feature extraction pattern resolution ensures that the directional derivative in the reverse gradient propagation is normalized near the branch point of the gradient propagation by making the difference distribution of the classifier in the form of cross entropy in relation to the weight matrix of the feature vector obtained after the expansion of the detection enhanced feature map and the reference enhanced feature map be consistent with the true feature difference distribution of the feature vector, that is, the gradient is weighted for the feature extraction patterns of the first convolutional neural network and the second convolutional neural network. Therefore, the resolution of the feature extraction mode is inhibited, the feature extraction capability of the first convolutional neural network on the image semantics of the plug state detection image and the feature extraction capability of the second convolutional neural network on the image semantics of the reference plug state image are improved, and the accuracy of the classification result of the classification feature map is correspondingly improved. Therefore, the plugging state of the current transformer can be effectively monitored in real time, an early warning prompt is generated when the plugging state of the current transformer is monitored to be abnormal, and then faults and accidents are avoided, so that the normal and safe operation of the ring network metering cabinet is ensured.
In summary, the ring network metering cabinet 100 according to the embodiment of the present application is clarified, where first, an obtained plug state detection image and a reference plug state image are passed through a twin network model including a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map, then, the detection feature map and the reference feature map are passed through a parallel weight distribution module respectively to obtain a detection enhanced feature map and a reference enhanced feature map, then, a difference feature map between the detection enhanced feature map and the reference enhanced feature map is calculated, and finally, the difference feature map is passed through a classifier to obtain a classification result for indicating whether the plug state of the metering current transformer meets a predetermined requirement. Through the mode, the plugging state of the current transformer can be effectively monitored in real time, so that faults and accidents are avoided, and the normal and safe work of the ring network metering cabinet is ensured.
As described above, the ring network metering cabinet 100 according to the embodiment of the present application may be implemented in various terminal devices, such as a server having a metering current transformer plug-in/pull-out state monitoring algorithm. In one example, the ring main unit 100 can be integrated into the terminal device as a software module and/or a hardware module. For example, the ring network metering cabinet 100 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the ring network metering cabinet 100 can also be one of many hardware modules of the terminal device.
Exemplary method
Fig. 6 illustrates a flowchart of a method for monitoring a plugging state of a metering current transformer according to an embodiment of the present application. As shown in fig. 6, the method for monitoring the plugging/unplugging state of a metering current transformer according to the embodiment of the present application includes: s110, acquiring a plug state detection image and a reference plug state image of the metering current transformer; s120, enabling the plugging and unplugging state detection image and the reference plugging and unplugging state image to pass through a twin network model comprising a first convolutional neural network and a second convolutional neural network so as to obtain a detection characteristic diagram and a reference characteristic diagram, wherein the first convolutional neural network and the second convolutional neural network have the same network structure; s130, respectively passing the detection feature map and the reference feature map through a parallel weight distribution module to obtain a detection enhancement feature map and a reference enhancement feature map; s140, calculating a difference feature map between the detection enhanced feature map and the reference enhanced feature map; and S150, the differential characteristic diagram is processed by a classifier to obtain a classification result, and the classification result is used for indicating whether the plugging state of the metering current transformer meets the preset requirement or not.
Fig. 7 illustrates a schematic diagram of a system architecture of a metering current transformer plugging status monitoring method according to an embodiment of the application. As shown in fig. 7, in the system architecture of the method for monitoring the plugging state of the metering current transformer, first, a plugging state detection image and a reference plugging state image of the metering current transformer are obtained; then, enabling the plugging and unplugging state detection image and the reference plugging and unplugging state image to pass through a twin network model comprising a first convolutional neural network and a second convolutional neural network to obtain a detection characteristic diagram and a reference characteristic diagram, wherein the first convolutional neural network and the second convolutional neural network have the same network structure; then, the detection feature map and the reference feature map are respectively passed through a parallel weight distribution module to obtain a detection enhancement feature map and a reference enhancement feature map; then, calculating a difference feature map between the detection enhanced feature map and the reference enhanced feature map; and finally, the differential characteristic diagram is subjected to a classifier to obtain a classification result, and the classification result is used for indicating whether the plugging state of the metering current transformer meets the preset requirement or not.
In a specific example, in the method for monitoring the plug state of the metering current transformer, the passing the plug state detection image and the reference plug state image through a twin network model including a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map includes: using each layer of the first convolutional neural network in the twin network model to respectively perform the following steps on input data in the forward direction transmission of the layer: performing convolution processing on the input data based on a two-dimensional convolution kernel to obtain a first convolution feature map; performing one-dimensional convolution processing on the input data based on a first one-dimensional convolution kernel to obtain a second convolution characteristic diagram; performing one-dimensional convolution processing on the input data based on a second one-dimensional convolution kernel to obtain a third convolution characteristic diagram; fusing the first convolution feature map, the second convolution feature map and the third convolution feature map to obtain a fused convolution feature map; pooling the fused convolution feature map to obtain a pooled feature map; carrying out nonlinear activation processing on the pooled feature map to obtain an activated feature map; the output of the last layer of the first convolutional neural network in the twin network model is the detection feature map, and the input of the first layer of the first convolutional neural network in the twin network model is the plugging and unplugging state detection image.
In a specific example, in the method for monitoring the plug state of the metering current transformer, the passing the plug state detection image and the reference plug state image through a twin network model including a first convolutional neural network and a second convolutional neural network to obtain a detection feature map and a reference feature map includes: performing, using layers of a second convolutional neural network in the twin network model, in forward pass of layers, input data respectively: performing convolution processing on the input data based on a two-dimensional convolution kernel to obtain a first convolution feature map; performing one-dimensional convolution processing on the input data based on a first one-dimensional convolution kernel to obtain a second convolution characteristic diagram; performing one-dimensional convolution processing on the input data based on a second one-dimensional convolution kernel to obtain a third convolution characteristic diagram; fusing the first convolution feature map, the second convolution feature map and the third convolution feature map to obtain a fused convolution feature map; pooling the fused convolution feature map to obtain a pooled feature map; carrying out nonlinear activation processing on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network in the twin network model is the reference characteristic diagram, and the input of the first layer of the second convolutional neural network in the twin network model is the reference plugging state image.
In a specific example, in the method for monitoring the plugging/unplugging state of the metering current transformer, the obtaining a detection enhanced feature map and a reference enhanced feature map by respectively passing through the detection feature map and the reference feature map by a parallel weight distribution module includes: respectively passing the detection feature map and the reference feature map through a spatial attention module of a parallel weight distribution module to obtain a detection spatial attention feature map and a reference spatial attention feature map; respectively passing the detection feature map and the reference feature map through a channel attention module of a parallel weight distribution module to obtain a detection channel attention feature map and a reference channel attention feature map; and fusing the detection space attention feature map and the detection channel attention feature map to obtain the detection enhancement feature map, and fusing the reference space attention feature map and the reference channel attention feature map to obtain the reference enhancement feature map.
In a specific example, in the method for monitoring the plugging/unplugging state of the metering current transformer, the step of respectively passing the detection feature map and the reference feature map through a spatial attention module of a parallel weight assignment module to obtain a detection spatial attention feature map and a reference spatial attention feature map further includes: depth convolution coding is carried out on the detection feature map and the reference feature map respectively by using a convolution coding part of a space attention module of the parallel weight distribution module so as to obtain a detection convolution feature map and a reference convolution feature map; inputting the detection convolution feature map and the reference convolution feature map into a spatial attention part of a spatial attention module of the parallel weight assignment module respectively to obtain a detection spatial attention map and a reference spatial attention map; respectively passing the detection space attention diagram and the reference space attention diagram through a Softmax activation function to obtain a detection space attention feature map and a reference space attention feature map; and calculating the point-by-point multiplication of the detection space attention feature map and the detection feature map to obtain the detection space attention feature map, and calculating the point-by-point multiplication of the reference space attention feature map and the reference feature map to obtain the reference space attention feature map.
In a specific example, in the method for monitoring the plugging/unplugging state of the metering current transformer, the obtaining the detection channel attention feature map and the reference channel attention feature map by respectively passing the detection feature map and the reference feature map through a channel attention module of a parallel weight assignment module further includes: inputting the detection feature map and the reference feature map into the multilayer convolution layer of the channel attention module of the parallel weight distribution module respectively to obtain a detection convolution feature map and a reference convolution feature map; respectively calculating the global mean value of each feature matrix of the detection convolution feature map and the reference convolution feature map along the channel dimension to obtain a detection channel feature vector and a reference channel feature vector; inputting the detection channel feature vector and the reference channel feature vector into the Sigmoid activation function respectively to obtain a detection channel attention weight vector and a reference channel attention weight vector; and weighting each feature matrix along the channel dimension of the detection convolution feature map by taking the feature value of each position in the detection channel attention weight vector as a weight to obtain the detection channel attention feature map, and weighting each feature matrix along the channel dimension of the reference convolution feature map by taking the feature value of each position in the reference channel attention weight vector as a weight to obtain the reference channel attention feature map.
In one specific example, the metering current transformer is plugged and unpluggedIn the state monitoring method, the calculating a difference feature map between the detection enhanced feature map and the reference enhanced feature map further includes: calculating a difference feature map between the detection enhancement feature map and the reference enhancement feature map in the following formula; wherein the formula is:
wherein, F 1 For said detection of enhanced feature maps, F 2 For the purpose of the reference enhancement feature map,
indicating a difference by position.
In a specific example, in the method for monitoring the plugging/unplugging state of the metering current transformer, the step of passing the differential feature map through a classifier to obtain a classification result further includes: processing the differential feature map by using the classifier according to the following formula to obtain the classification result, wherein the formula is as follows: o = softmax { (W) n ,B n ):…:(W 1 ,B 1 )|Project(F d ) Wherein, project (F) d ) Representing the projection of the difference profile as a vector, W 1 To W n As a weight matrix for each fully connected layer, B 1 To B n A bias matrix representing the layers of the fully connected layer.
In a specific example, in the method for monitoring the plugging/unplugging state of the metering current transformer, the method further includes: training the twin network model, the parallel weight assignment module and the classifier; wherein the training the twin network model, the parallel weight assignment module, and the classifier comprises: acquiring training data, wherein the training data comprises a training plug state detection image and a training reference plug state image of the metering current transformer, and whether the plug state of the metering current transformer meets a true value of a preset requirement or not; enabling the training plug state detection image and the training reference plug state image to pass through the twin network model comprising the first convolutional neural network and the second convolutional neural network to obtain a training detection feature map and a training reference feature map, wherein the first convolutional neural network and the second convolutional neural network have the same network structure; respectively passing the training detection feature map and the training reference feature map through the parallel weight distribution module to obtain a training detection enhanced feature map and a training reference enhanced feature map; calculating a training difference feature map between the training detection enhanced feature map and the training reference enhanced feature map; passing the training difference feature map through the classifier to obtain a classification loss function value; calculating inhibition loss function values of feature extraction mode resolution of the training detection enhanced feature map and the training reference enhanced feature map; and training the twin network model, the parallel weight distribution module and the classifier by taking the weighted sum of the suppression loss function value and the classification loss function value of the feature extraction mode resolution as a loss function value.
In a specific example, in the method for monitoring the plugging/unplugging state of the metering current transformer, the calculating a suppression loss function value of the feature extraction mode resolution of the training detection enhanced feature map and the training reference enhanced feature map further includes: calculating a suppression loss function value resolved by the feature extraction mode of the training detection enhancement feature map and the training reference enhancement feature map according to the following formula; wherein the formula is:
wherein V 1 And V 2 Respectively, the feature vectors obtained after the detection enhanced feature map and the reference enhanced feature map are expanded, and M is 1 And M 2 The weights of the classifier on the feature vectors obtained after the detection enhanced feature map and the reference enhanced feature map are expandedThe matrix is a matrix of a plurality of matrices,
represents the square of the two-norm of the vector, | - | F The F-norm of the matrix is represented,
representing a position-wise subtraction and log represents a base-2 logarithmic function.
Here, it can be understood by those skilled in the art that the specific operations of the steps in the above-mentioned method for monitoring the plugging and unplugging state of the metering current transformer have been described in detail in the above description of the ring network metering cabinet with reference to fig. 1 to 5, and therefore, the repeated description thereof will be omitted.
Claims (10)
1. A looped network measurement cabinet, its characterized in that includes:
the plug state monitoring unit is used for acquiring a plug state detection image and a reference plug state image of the metering current transformer;
the plug state encoding unit is used for enabling the plug state detection image and the reference plug state image to pass through a twin network model comprising a first convolutional neural network and a second convolutional neural network so as to obtain a detection characteristic diagram and a reference characteristic diagram, wherein the first convolutional neural network and the second convolutional neural network have the same network structure;
the feature enhancement unit is used for enabling the detection feature map and the reference feature map to pass through a parallel weight distribution module respectively so as to obtain a detection enhancement feature map and a reference enhancement feature map;
a difference characterization unit, configured to calculate a difference feature map between the detected enhanced feature map and the reference enhanced feature map; and the monitoring result generating unit is used for enabling the differential characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the plugging state of the metering current transformer meets the preset requirement or not.
2. The ring network metering cabinet of claim 1, wherein the plug status encoding unit comprises:
a detection image coding subunit, configured to perform, in a layer forward pass, the following operations on the input data respectively using each layer of the first convolutional neural network in the twin network model:
performing convolution processing on the input data based on a two-dimensional convolution kernel to obtain a first convolution feature map;
performing one-dimensional convolution processing on the input data based on a first one-dimensional convolution kernel to obtain a second convolution characteristic diagram;
performing one-dimensional convolution processing on the input data based on a second one-dimensional convolution kernel to obtain a third convolution characteristic diagram;
fusing the first convolution feature map, the second convolution feature map and the third convolution feature map to obtain a fused convolution feature map;
pooling the fused convolution feature map to obtain a pooled feature map; performing nonlinear activation processing on the pooled feature map to obtain an activated feature map;
the output of the last layer of the first convolutional neural network in the twin network model is the detection feature map, and the input of the first layer of the first convolutional neural network in the twin network model is the plugging and unplugging state detection image.
3. The ring network metering cabinet of claim 2, wherein the plug status encoding unit comprises:
a reference image encoding subunit, configured to perform, in a layer forward pass, the following on the input data respectively using layers of a second convolutional neural network in the twin network model:
performing convolution processing on the input data based on a two-dimensional convolution kernel to obtain a first convolution feature map;
performing one-dimensional convolution processing on the input data based on a first one-dimensional convolution kernel to obtain a second convolution characteristic diagram;
performing one-dimensional convolution processing on the input data based on a second one-dimensional convolution kernel to obtain a third convolution characteristic diagram;
fusing the first convolution feature map, the second convolution feature map and the third convolution feature map to obtain a fused convolution feature map;
pooling the fused convolution feature map to obtain a pooled feature map; performing nonlinear activation processing on the pooled feature map to obtain an activated feature map;
and the output of the last layer of the second convolutional neural network in the twin network model is the reference characteristic diagram, and the input of the first layer of the second convolutional neural network in the twin network model is the reference plugging state image.
4. The ring main unit as claimed in claim 3, wherein the feature enhancing unit comprises:
the spatial attention branching subunit is used for respectively passing the detection feature map and the reference feature map through a spatial attention module of the parallel weight distribution module to obtain a detection spatial attention feature map and a reference spatial attention feature map;
the channel attention branching subunit is used for enabling the detection feature map and the reference feature map to pass through a channel attention module of the parallel weight distribution module respectively to obtain a detection channel attention feature map and a reference channel attention feature map; and a fusion subunit, configured to fuse the detection spatial attention feature map and the detection channel attention feature map to obtain the detection enhancement feature map, and fuse the reference spatial attention feature map and the reference channel attention feature map to obtain the reference enhancement feature map.
5. The ring main unit as claimed in claim 4, wherein the spatial attention branch subunit is further configured to:
depth convolution coding is carried out on the detection feature map and the reference feature map respectively by using a convolution coding part of a space attention module of the parallel weight distribution module so as to obtain a detection convolution feature map and a reference convolution feature map;
inputting the detection convolution feature map and the reference convolution feature map into a spatial attention part of a spatial attention module of the parallel weight assignment module respectively to obtain a detection spatial attention map and a reference spatial attention map;
respectively activating functions of the detection space attention diagram and the reference space attention diagram through Softmax to obtain a detection space attention feature map and a reference space attention feature map; and calculating the detection space attention feature map and multiplying the detection space attention feature map by the position points to obtain the detection space attention feature map, and calculating the reference space attention feature map and multiplying the reference space attention feature map by the position points to obtain the reference space attention feature map.
6. The ring main unit as claimed in claim 5, wherein the channel attention branch subunit is further configured to:
inputting the detection feature map and the reference feature map into the multilayer convolution layer of the channel attention module of the parallel weight distribution module respectively to obtain a detection convolution feature map and a reference convolution feature map;
respectively calculating the global mean value of each feature matrix of the detection convolution feature map and the reference convolution feature map along the channel dimension to obtain a detection channel feature vector and a reference channel feature vector;
inputting the detection channel feature vector and the reference channel feature vector into the Sigmoid activation function respectively to obtain a detection channel attention weight vector and a reference channel attention weight vector; and weighting each feature matrix along the channel dimension of the detection convolution feature map by taking the feature value of each position in the detection channel attention weight vector as a weight to obtain the detection channel attention feature map, and weighting each feature matrix along the channel dimension of the reference convolution feature map by taking the feature value of each position in the reference channel attention weight vector as a weight to obtain the reference channel attention feature map.
7. The ring main unit as claimed in claim 6, wherein the difference characterization unit is further configured to: calculating a difference feature map between the detection enhancement feature map and the reference enhancement feature map in the following formula;
wherein the formula is:
wherein, F 1 For said detection of enhanced feature maps, F 2 For the purpose of the reference enhancement feature map,
indicating a difference by position.
8. The ring network metering cabinet of claim 7, wherein the monitoring result generating unit is further configured to: processing the differential feature map using the classifier in the following formula to obtain the classification result,
wherein the formula is: o = softmax { (W) n ,B n ):…:(W 1 ,B 1 )|Project(F d ) Wherein, project (F) d ) Representing the projection of the difference profile as a vector, W 1 To W n As a weight matrix for each fully connected layer, B 1 To B n A bias matrix representing the layers of the fully connected layer.
9. The ring main unit as claimed in claim 8, further comprising a training module for training the twin network model, the parallel weight assignment module and the classifier;
wherein the training module comprises:
the training plug-in state monitoring unit is used for acquiring training data, wherein the training data comprise a training plug-in state detection image and a training reference plug-in state image of the metering current transformer, and whether the plug-in state of the metering current transformer meets a true value of a preset requirement or not;
a training plug state encoding unit, configured to pass the training plug state detection image and the training reference plug state image through the twin network model including the first convolutional neural network and the second convolutional neural network to obtain a training detection feature map and a training reference feature map, where the first convolutional neural network and the second convolutional neural network have the same network structure;
the training feature enhancement unit is used for enabling the training detection feature map and the training reference feature map to pass through the parallel weight distribution module respectively so as to obtain a training detection enhancement feature map and a training reference enhancement feature map;
the training difference characterization unit is used for calculating a training difference feature map between the training detection enhanced feature map and the training reference enhanced feature map;
the classification loss unit is used for enabling the training difference characteristic diagram to pass through the classifier to obtain a classification loss function value;
the inhibition loss function value calculation unit is used for calculating inhibition loss function values of feature extraction mode resolution of the training detection enhanced feature map and the training reference enhanced feature map; and the training unit is used for training the twin network model, the parallel weight distribution module and the classifier by taking the weighted sum of the inhibition loss function value and the classification loss function value of the feature extraction mode resolution as a loss function value.
10. The ring main unit of claim 9, wherein the mitigation loss function value calculation unit is further configured to: calculating a suppression loss function value resolved by the feature extraction mode of the training detection enhancement feature map and the training reference enhancement feature map according to the following formula;
wherein the formula is:
wherein V 1 And V 2 Respectively, the feature vectors obtained after the detection enhanced feature map and the reference enhanced feature map are expanded, and M is 1 And M 2 Respectively, the classifier is used for obtaining a weight matrix of feature vectors after the detection enhanced feature map and the reference enhanced feature map are developed,
represents the square of the two-norm of the vector, | - | F The F-norm of the matrix is represented,
representing a position-wise subtraction and log represents a base-2 logarithmic function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211337909.7A CN115731455A (en) | 2022-10-28 | 2022-10-28 | Looped network metering cabinet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211337909.7A CN115731455A (en) | 2022-10-28 | 2022-10-28 | Looped network metering cabinet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115731455A true CN115731455A (en) | 2023-03-03 |
Family
ID=85295003
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211337909.7A Pending CN115731455A (en) | 2022-10-28 | 2022-10-28 | Looped network metering cabinet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115731455A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116295116A (en) * | 2023-04-13 | 2023-06-23 | 广东省旭晟半导体股份有限公司 | Infrared emission module and preparation method thereof |
-
2022
- 2022-10-28 CN CN202211337909.7A patent/CN115731455A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116295116A (en) * | 2023-04-13 | 2023-06-23 | 广东省旭晟半导体股份有限公司 | Infrared emission module and preparation method thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP1714366B1 (en) | High impedance fault detection | |
| Judd et al. | Partial discharge monitoring of power transformers using UHF sensors. Part I: sensors and signal interpretation | |
| CN101553738B (en) | Flashover analysis tool | |
| CN103698653A (en) | System and method for detecting earth fault of sheath of single-core power cable on basis of circulation measurement | |
| US20080033602A1 (en) | Method and Device for Detecting Electric Arc Phenomenon on at Least One Electric Cable | |
| CN105223427B (en) | The detection method and detection device of fault electric arc | |
| CN105703258B (en) | GIS switch equipment action state monitoring system and use method thereof | |
| CN203811728U (en) | Single-core power cable sheath earth fault detecting system based on circulation measurement | |
| Lanzrath et al. | HPEM-based risk assessment of substations enabled for the smart grid | |
| CN205622133U (en) | GIS switchgear action state monitoring system | |
| Andrea et al. | Arcing fault in aircraft distribution network | |
| Ashok et al. | Fault diagnosis scheme for cross-country faults in dual-circuit line with emphasis on high-impedance fault syndrome | |
| CN115731455A (en) | Looped network metering cabinet | |
| CN109596956B (en) | DC series arc detection method and device | |
| CN109342882A (en) | Cable fault on-line monitoring method based on multi-source Eigenvalues analysis | |
| Radhakrishnan et al. | Synchrophasor based islanding detection for microgrids using moving window principal component analysis and extended mathematical morphology | |
| CN105301426B (en) | A kind of multi-layer selectivity low-voltage short-circuiting Protection system | |
| CN106530182A (en) | Protective method for electrical fire hazards based on three-proofing means | |
| Hong et al. | Detection of open conductor fault using multiple measurement factors of feeder RTUs in power distribution networks with DGs | |
| CN109522615A (en) | The zero sequence of relay protection of power system is shoved tolerance appraisal procedure and device | |
| CN112924813A (en) | Power distribution network short-circuit fault monitoring method and device based on electrical data | |
| CN107632237A (en) | A kind of fault line selection method for single-phase-to-ground fault based on wavelet character amount correlation after failure | |
| CN114924207B (en) | Method for detecting intermittent ground fault based on machine vision | |
| CN114283330B (en) | Online inspection identification method and system based on multi-source data | |
| CN115395660A (en) | Distribution transformer flood disaster satellite image monitoring method and system |
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 |