Ultrahigh voltage equipment maintenance skill behavior identification method and system

Abstract

The invention discloses an ultrahigh-voltage equipment maintenance skill behavior identification method and system, and relates to the technical field of power equipment maintenance, and the method comprises the following steps: carrying out the layering of ultrahigh-voltage maintenance demonstration operation according to first data, obtaining a maintenance operation layer, and enabling the first data to be obtained through the assessment of the maintenance operation relevance; performing information entropy analysis on the maintenance demonstration operation to obtain second data of each maintenance operation layer, wherein the second data comprises skeleton joint criticality; obtaining a multi-source video stream of real maintenance operation, and adjusting the second data according to the misoperation action to obtain dynamic second data; and identifying the real maintenance operation based on ST-GCN according to a second skeleton feature, wherein the second skeleton feature is obtained by processing the obtained first skeleton feature in combination with dynamic second data. The method is used for solving the problems that in an ultrahigh-voltage equipment maintenance scene, structural representation of a complex operation process is difficult, and key dynamic characteristics are not accurately captured by a behavior recognition model.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment maintenance, and more specifically, to a method and system for identifying the maintenance skill behaviors of extra-high voltage equipment. Background Art

[0002] Extra-high voltage equipment is crucial in the power system, and its maintenance work is of great significance. Traditional planned power outage maintenance can no longer meet the development needs of the power industry, and condition-based maintenance has become the main development direction. Condition-based maintenance is to observe, analyze, and evaluate the operating state of extra-high voltage equipment, and perform targeted inspections and repairs based on the results, which has the advantages of clear maintenance purpose, reduced capital investment, and reduced maintenance workload. In actual operation, there are various maintenance methods. For example, the acoustic fingerprint inspection technology is applied. By using an unmanned aerial vehicle (UAV) equipped with an acoustic fingerprint device and a camera, real-time monitoring and diagnosis of the insulator string of the transmission line are carried out. Another example is to carry out the full-process intelligent maintenance of the transmission line. By using multiple UAVs equipped with special intelligent devices to complete multiple tasks such as flaw detection and voltage detection, and the data is uploaded and analyzed in real time. In addition, new construction methods and tooling are developed to improve the maintenance efficiency and safety of extra-high voltage large-scale substation equipment. For example, maintenance platforms, constant temperature devices, lifting tooling, etc. are developed, and the detection method for the transformer without disconnecting the leads is determined through experiments.

[0003] In power grid enterprises, it is extremely important to identify abnormal behaviors during the maintenance process of substation equipment. The traditional manual inspection method is cumbersome and time-consuming, resulting in low efficiency in identifying abnormal behaviors. With the development of technology, automatic identification methods using technologies such as image recognition have gradually emerged. For example, by inputting the image data of the equipment to be maintained into a pre-trained maintenance analysis model for feature extraction, and then identifying the abnormal behaviors based on the generated auxiliary information for identifying abnormal behaviors, so as to improve the identification efficiency. At the same time, some intelligent platforms and devices are also applied to the identification of maintenance behaviors.

[0004] For example, a method and related device for detecting faults in a distribution network based on a digital twin system disclosed in the invention patent announcement with the publication number of CN116298701B. A method and related device for detecting faults in a distribution network based on a digital twin system are provided, which relates to the technical field of power grid transmission and distribution fault monitoring. The method includes: constructing a three-dimensional model of the distribution network according to the preset distribution network entity data and the power scene image, and compiling the Beidou grid code; fusing the real-time distribution network data information to obtain the digital twin body of the distribution network; when a fault is detected in the digital twin body, obtaining the maintenance image of the faulty power equipment through the distribution network camera, and identifying the Beidou grid code where the fault is located; the digital twin body can accurately reflect the state and information of the distribution network in a timely manner, and accurately locate the power equipment with the Beidou grid code. When a faulty device appears, the Beidou grid code where the fault is located is identified through the corresponding camera, so that the fault can be quickly and accurately repaired, and the stable power supply of the distribution network is guaranteed.

[0005] For example, the interactive behavior recognition method and system based on multi-modal group behavior recognition technology announced in the invention patent announcement with the announcement number of CN119296183B discloses an interactive behavior recognition method and system based on multi-modal group behavior recognition technology, and proposes an interactive behavior recognition method and system based on multi-modal group behavior recognition technology. By designing a multi-modal group behavior recognition network, including an image branch sub-network, a text branch sub-network, and a key point branch sub-network, multi-modal feature processing is carried out respectively to avoid the problem of low recognition accuracy caused by factors such as interactive behaviors between groups and complex interactive environments when recognizing through single-modal information, further conforming to the actual maintenance situation, improving the overall recognition accuracy, and through lightweight design, improving the recognition efficiency. Then, the relationships and interactive enhancements between groups and between groups and interactive targets are carried out, further improving the recognition accuracy and avoiding recognition obstacles caused by occlusion and overlap. The present invention improves the recognition accuracy and efficiency of the interactive behaviors of group targets.

[0006] In the above disclosed technical solution, at least the following technical problems exist: There are two core defects in the existing technology when dealing with the recognition of ultra-high voltage equipment maintenance behaviors: on the one hand, traditional methods use single-time series modeling or fixed process decomposition, which are difficult to effectively capture composite features such as semantic relevance, action coherence, and pause duration in multi-step maintenance operations, resulting in insufficient structured representation ability for complex operation processes and being unable to support refined hierarchical analysis; on the other hand, when the behavior recognition model based on general scenarios is directly applied to the high-voltage maintenance scenario, due to the lack of a dynamic weight adjustment mechanism guided by domain knowledge, the weights of its static skeleton joint features are easily affected by environmental interference and individual differences, and the sensitivity to the subtle dynamic changes of key skeleton joints in misoperation actions is insufficient.

[0007] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0008] In order to overcome the above defects of the existing technology, an embodiment of the present invention provides an ultra-high voltage equipment maintenance skill behavior recognition method and system, by constructing a hierarchical maintenance operation hierarchical architecture based on semantic association and integrating a dynamic skeleton joint key degree adjustment mechanism of information entropy analysis and misoperation feedback to solve the problems of difficult structured representation of complex operation processes in the ultra-high voltage equipment maintenance scenario and inaccurate capture of key dynamic features by the behavior recognition model.

[0009] To achieve the above object, the present invention provides the following technical solutions: A method for identifying the skills and behaviors of ultra-high voltage equipment maintenance, comprising the following steps: In a preferred embodiment, the ultra-high voltage maintenance demonstration operations are stratified according to the first data to obtain maintenance operation layers, where the first data is obtained by evaluating the relevance of maintenance operations; the second data of each maintenance operation layer is obtained by analyzing the maintenance demonstration operations based on information entropy, and the second data includes the key degrees of skeleton joints; a multi-source video stream of the actual maintenance operations is acquired, and the second data is adjusted according to the misoperation actions to obtain dynamic second data; the actual maintenance operations are identified based on ST-GCN according to the second skeleton features, and the second skeleton features are obtained by processing the acquired first skeleton features in combination with the dynamic second data.

[0010] In a preferred embodiment, the method for obtaining the first data is specifically as follows: The ultra-high voltage maintenance operations are disassembled into a set of operation units; a correlation matrix is constructed according to the semantic correlation degree, coherence score, and pause duration between the operation units, where the semantic correlation degree is obtained based on the power maintenance knowledge graph, the coherence score is quantitatively evaluated by expert scoring to obtain the action coherence of the operation units, and the pause duration is obtained by averaging the pause durations of each operation unit; hierarchical clustering analysis is performed on the correlation matrix to obtain the first data.

[0011] In a preferred embodiment, the method for obtaining the second data is specifically as follows: Determine the human body parts involved in each operation unit, and preset the participation weights of each part in each operation unit; obtain the entropy difference before and after the maintenance actions of each operation unit within a single maintenance operation layer; according to the participation weights and entropy differences, obtain the operation key degrees of each part in a single maintenance action; accumulate the operation key degrees of each part in a single maintenance action to obtain the second data within a single maintenance operation layer.

[0012] In a preferred embodiment, the method for obtaining the dynamic second data is specifically as follows: Based on expert opinions, obtain the severity of the impact of each part in the misoperation actions on the maintenance operation layer where it is located; weight the second data within a single maintenance operation layer according to the severity and normalize it to obtain the dynamic second data.

[0013] In a preferred embodiment, the skeleton joints are obtained by mapping the human body parts involved in the maintenance actions and the obtained physiological rotation centers.

[0014] In a preferred embodiment, the processing of the first skeleton features is specifically as follows: Align the first skeleton features with the dynamic second data; multiply the aligned dynamic second data and the first skeleton features element by element to obtain the second skeleton features.

[0015] In a preferred embodiment, the recognition of the actual maintenance operation based on ST-GCN is specifically as follows: a behavior recognition model is constructed based on ST-GCN, and the second skeleton feature is used as the input feature; the maintenance operation is recognized according to the behavior recognition model.

[0016] In a preferred embodiment, the mapping of the human body parts participating in the maintenance action and the obtained physiological rotation center is as follows: the central position of the human body parts participating in the maintenance action is marked as the first self-loop, and the obtained physiological rotation center is marked as the second self-loop. The self-loop is used to represent the skeleton joint, and the first self-loop is distinguished from the second self-loop by color features; and a bidirectional edge is connected between two self-loops with physical associations.

[0017] A high-voltage equipment maintenance skill behavior recognition system includes: a maintenance operation stratification module, configured to stratify the high-voltage maintenance demonstration operation according to the first data to obtain a maintenance operation layer, where the first data is obtained by evaluating the relevance of the maintenance operation; a skeleton joint key degree analysis module, configured to obtain second data of each maintenance operation layer based on information entropy analysis of the maintenance demonstration operation, where the second data includes the skeleton joint key degree; a real-time adjustment module, configured to obtain a multi-source video stream of the actual maintenance operation, and adjust the second data according to the misoperation action to obtain dynamic second data; an operation recognition module, configured to recognize the actual maintenance operation based on ST-GCN according to the second skeleton feature, where the second skeleton feature is obtained by processing the obtained first skeleton feature by combining the dynamic second data.

[0018] The technical effects and advantages of the high-voltage equipment maintenance skill behavior recognition method and system of the present invention: The present invention hierarchically clusters and stratifies the high-voltage maintenance operation based on semantic correlation degree, coherence score, and pause duration to construct an interpretable maintenance operation layer structure; innovatively introduces information entropy analysis combined with human body part weights to quantify the dynamic key degree of the skeleton joint, and performs key degree adaptive adjustment based on the misoperation video stream to form a dynamic weight mechanism integrating domain knowledge; further uses the graph theory method to construct a skeleton node model with physical association characteristics, and realizes feature fusion with enhanced key degree through ST-GCN, and finally constructs a behavior recognition system with a three-layer optimization mechanism. Its core advantages are reflected in: significantly improving the structured representation ability of complex maintenance processes through operation stratification, and realizing feature adaptive enhancement in misoperation scenarios by using the dynamic key degree mechanism, which is particularly suitable for high-risk power maintenance scenarios with subtle operation differences. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic flow chart of a high-voltage equipment maintenance skill behavior recognition method provided by an embodiment of the present invention.

[0020] Figure 2 This is a schematic structural diagram of a system for identifying the maintenance skill behaviors of ultra-high voltage equipment provided by an embodiment of the present invention. Specific implementation manners

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Embodiment 1 Figure 1 A method and system for identifying the maintenance skill behaviors of ultra-high voltage equipment of the present invention are provided, including the following steps: S1. According to the first data, the ultra-high voltage maintenance demonstration operations are stratified to obtain a maintenance operation layer, and the first data is obtained by evaluating the relevance of the maintenance operations; S2. The second data of each maintenance operation layer is obtained by analyzing the maintenance demonstration operations based on information entropy, and the second data includes the key degree of the skeleton joints; S3. A multi-source video stream of the actual maintenance operation is obtained, and the second data is adjusted according to the incorrect operation actions to obtain dynamic second data; S4. The actual maintenance operation is identified based on the second skeleton feature by using ST-GCN, and the second skeleton feature is obtained by processing the acquired first skeleton feature in combination with the dynamic second data.

[0023] In this embodiment, the ultra-high voltage maintenance operations are hierarchically clustered and stratified based on semantic relevance, coherence score, and pause duration to construct an interpretable maintenance operation layer structure; information entropy analysis is innovatively introduced and combined with the weights of human body parts to quantify the dynamic key degree of the skeleton joints, and the key degree is adaptively adjusted based on the incorrect operation video stream to form a dynamic weight mechanism integrating domain knowledge; further, a skeleton node model with physical association characteristics is constructed by using the graph theory method, and feature fusion with enhanced key degree is realized through ST-GCN. Finally, a behavior recognition system with a three-layer optimization mechanism is constructed. Its core advantages are reflected in: significantly improving the structured representation ability of complex maintenance processes through operation stratification, realizing feature self-adaptation enhancement in incorrect operation scenarios by using the dynamic key degree mechanism, and being particularly suitable for high-risk power maintenance scenarios with subtle operation differences.

[0024] S1. According to the first data, the ultra-high voltage maintenance demonstration operations are stratified to obtain a maintenance operation layer, and the first data is obtained by evaluating the relevance of the maintenance operations.

[0025] In this embodiment, the method for obtaining the first data is specifically as follows: Decompose the ultra-high voltage maintenance operation into a set of operation units; Construct a correlation matrix based on the semantic correlation degree, coherence score, and pause duration among the operation units. The semantic correlation degree is obtained based on the power maintenance knowledge graph. The coherence score is quantitatively evaluated by expert scoring to obtain the action coherence of the operation units, and the pause duration is obtained by averaging the pause durations of each operation unit; Perform hierarchical clustering analysis on the correlation matrix to obtain the first data.

[0026] In this embodiment, the ultra-high voltage maintenance demonstration operation is decomposed into a set of operation units, and the decomposition principles include independence and indivisibility.

[0027] It should be noted that independence means that an operation unit does not depend on the results of other units, and indivisibility means that the actions within an operation unit have physical continuity.

[0028] It should be noted that averaging the pause durations of each operation unit means taking the average of the pause durations between the same operation units in multiple ultra-high voltage maintenance operation videos.

[0029] In this embodiment, the specific formula for the coherence score is:

[0030] In the formula, is the coherence score between the i-th operation unit and the j-th operation unit, is the score given by the m-th expert for the action coherence of the operation unit, is the total number of experts.

[0031] In this embodiment, the specific formula for the correlation matrix is:

[0032] In the formula, is the correlation degree between the i-th operation unit and the j-th operation unit, is the semantic correlation degree between the i-th operation unit and the j-th operation unit obtained based on the knowledge graph, is the coherence score between the i-th operation unit and the j-th operation unit, is the highest preset coherence score, is the pause duration between the i-th operation unit and the j-th operation unit after normalization processing, 、 and are weight coefficients.

[0033] S2. Analyze the maintenance demonstration operation based on information entropy to obtain the second data of each maintenance operation layer. The second data includes the key degree of the skeleton joints.

[0034] In this embodiment, the method for obtaining the second data is specifically as follows: Determine the human body parts involved in each operation unit, and preset the participation weights of each part in each operation unit; Obtain the entropy difference before and after the maintenance actions of each operation unit within a single maintenance operation layer; Based on the participation weight and the entropy difference, obtain the operation criticality of each part in a single maintenance action; Accumulate the operation criticality of each part in a single maintenance action to obtain the second data within a single maintenance operation layer.

[0035] In this embodiment, the specific calculation formula for the operation criticality of each part in a single maintenance action is:

[0036] In the formula, is the operation criticality of the human body part k in the i-th operation unit, is the entropy difference before and after the maintenance action of the i-th operation unit, is the participation weight of the human body part k in the i-th operation unit.

[0037] In this embodiment, the specific calculation formula for the second data within a single maintenance operation layer is:

[0038] In the formula, is the skeleton joint criticality of the human body part k within a single maintenance operation layer, is the total number of operation units within the maintenance operation layer, is the operation criticality of the human body part k in the i-th operation unit.

[0039] In this embodiment, the skeleton joint is obtained by mapping the human body parts involved in the maintenance action and the obtained physiological rotation center.

[0040] In this embodiment, the mapping of the human body parts involved in the maintenance action and the obtained physiological rotation center, the specific mapping method is: Mark the center position of the human body parts involved in the maintenance action as the first self-loop, and mark the obtained physiological rotation center as the second self-loop. The self-loop is used to represent the skeleton joint, and the first self-loop is distinguished from the second self-loop by color features; And connect the two self-loops with a physical association by a bidirectional edge.

[0041] S3. Obtain the multi-source video stream of the actual maintenance operation, and adjust the second data according to the misoperation actions to obtain the dynamic second data.

[0042] In this embodiment, the method for obtaining the dynamic second data is specifically as follows: Based on expert opinions, obtain the severity of the impact of each part in the misoperation action on the current maintenance operation layer. Weight the second data within a single maintenance operation layer according to the severity and normalize it to obtain the dynamic second data.

[0043] In this embodiment, the value range of the severity is:

[0044] In the formula, is the severity of the misoperation of part k.

[0045] S4. Based on the second skeleton feature, identify the true maintenance operation by using ST-GCN, where the second skeleton feature is obtained by processing the acquired first skeleton feature by combining the dynamic second data.

[0046] In this embodiment, the processing of the first skeleton feature is specifically as follows: Align the first skeleton feature with the dynamic second data; Multiply the aligned dynamic second data and the first skeleton feature element by element to obtain the second skeleton feature.

[0047] In this embodiment, the identification of the true maintenance operation by using ST-GCN is specifically as follows: Build a behavior recognition model based on ST-GCN and use the second skeleton feature as the input feature; Identify the maintenance operation according to the behavior recognition model.

[0048] In this embodiment, obtain real-time inference of the multi-source camera video stream, comprehensively judge and process the recognition results in the post-processing link, and send the key frame information in the detection results to the background for storage together.

[0049] It should be noted that the first skeleton feature refers to the original skeleton feature extracted from the multi-source video stream of the true maintenance operation.

[0050] Spatial-Temporal Graph Convolutional Network (ST-GCN) is used to solve the problem of human action recognition based on human skeleton key points. It takes human key points as nodes and constructs a spatio-temporal graph with human structure and time as graph edges, extending the graph convolutional neural network to a spatio-temporal graph model. It captures the mutual relationship between joints through convolution in the spatial dimension and understands the sequential changes of actions through convolution in the temporal dimension, being able to consider both human structure and action sequence information, thus more accurately recognizing complex human actions. The input of this model is usually skeleton-based data, which can be obtained from motion capture devices or videos through pose estimation algorithms, and the data dimension is generally (N, C, T, V, M). ST-GCN has the advantages of being lightweight and scalable.

[0051] Embodiment 2, Figure 2 This invention provides a system for recognizing the skills and behaviors of ultra-high voltage equipment maintenance, including: A maintenance operation stratification module, which is used to stratify the ultra-high voltage maintenance demonstration operations according to the first data to obtain the maintenance operation layers, and the first data is obtained by evaluating the relevance of maintenance operations; A skeleton joint key degree analysis module, which is used to analyze the maintenance demonstration operations based on information entropy to obtain the second data of each maintenance operation layer, and the second data includes the skeleton joint key degree; A real-time adjustment module, which is used to obtain the multi-source video stream of the actual maintenance operation, and adjust the second data according to the misoperation actions to obtain the dynamic second data; An operation recognition module, which is used to recognize the actual maintenance operation based on the second skeleton feature by using ST-GCN, and the second skeleton feature is obtained by processing the acquired first skeleton feature by combining the dynamic second data.

[0052] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation, and the preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0053] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0054] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0055] In addition, each functional module in various embodiments of the present application may be integrated into a processing module, may exist physically alone for each module, or two or more modules may be integrated into one module.

[0056] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claimed rights.

[0057] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included within the protection scope of the present invention.

Claims

1. A method for identifying the maintenance skill behaviors of ultra-high voltage equipment, characterized in that, Including the following steps: Stratify the ultra-high voltage maintenance demonstration operation according to the first data to obtain a maintenance operation layer, where the first data is obtained by evaluating the relevance of maintenance operations; Perform information entropy analysis on the maintenance demonstration operation to obtain the second data of each maintenance operation layer, where the second data includes the key degree of skeleton joints; Obtain the multi-source video stream of the actual maintenance operation, and adjust the second data according to the misoperation actions to obtain the dynamic second data; Identify the actual maintenance operation based on ST-GCN according to the second skeleton feature, where the second skeleton feature is obtained by processing the acquired first skeleton feature in combination with the dynamic second data.

2. The method for identifying the overhaul skill behaviors of the ultra-high voltage equipment according to claim 1, wherein, The method for obtaining the first data is specifically as follows: Decompose the ultra-high voltage maintenance operation into a set of operation units; Construct a correlation matrix according to the semantic correlation degree, coherence score, and pause duration between operation units. The semantic correlation degree is obtained based on the power maintenance knowledge graph, the coherence score is obtained by quantitatively evaluating the action coherence of operation units through expert scoring, and the pause duration is obtained by averaging the pause durations of each operation unit; Perform hierarchical clustering analysis on the correlation matrix to obtain the first data.

3. The method for identifying the skills and behaviors of overhauling ultra-high voltage equipment according to claim 2, wherein The method for obtaining the second data is specifically as follows: Determine the human body parts involved in each operation unit, and preset the participation weights of each part in each operation unit; Obtain the entropy difference before and after the maintenance action of each operation unit within a single maintenance operation layer; According to the participation weight and entropy difference, obtain the operation key degree of each part in a single maintenance action; Accumulate the operation key degrees of each part in a single maintenance action to obtain the second data within a single maintenance operation layer.

4. The method for identifying the maintenance skill behaviors of the ultra-high voltage equipment according to claim 3, wherein, The method for obtaining the dynamic second data is specifically as follows: Based on expert opinions, obtain the severity of the impact of each part in the misoperation action on the maintenance operation layer where it is located; Weight the second data within a single maintenance operation layer according to the severity and normalize it to obtain the dynamic second data.

5. The method for identifying the skills and behaviors of overhauling ultra-high voltage equipment according to claim 4, characterized in that, The skeleton joint is obtained by mapping the human body parts involved in the maintenance action and the acquired physiological rotation center.

6. The method for identifying the skills and behaviors of ultra-high voltage equipment maintenance according to claim 5, characterized in that, The processing of the first skeleton feature is specifically as follows: Align the first skeleton feature with the dynamic second data; Multiply the aligned dynamic second data and the first skeleton feature element by element to obtain the second skeleton feature.

7. The method for identifying the skills and behaviors of ultra-high voltage equipment maintenance according to claim 6, wherein The identification of the actual maintenance operation based on ST-GCN is specifically as follows: Construct a behavior recognition model based on ST-GCN and use the second skeleton feature as the input feature; Identify the maintenance operation according to the behavior recognition model.

8. The method for identifying the skills and behaviors of overhauling ultra-high voltage equipment according to claim 7, wherein, The mapping of the human body parts involved in the maintenance action and the acquired physiological rotation center, and the specific mapping method is as follows: Mark the center position of the human body parts involved in the maintenance action as the first self-loop, and mark the acquired physiological rotation center as the second self-loop. The self-loop is used to represent the skeleton joint, and the first self-loop is distinguished from the second self-loop by color features; Connect two self-loops with physical associations with a bidirectional edge.

9. A system using the method for identifying ultra-high voltage equipment maintenance skill behaviors described in any one of claims 1-8, including: The maintenance operation layering module is used to layer the ultra-high voltage maintenance demonstration operation according to the first data to obtain the maintenance operation layer, where the first data is obtained by evaluating the relevance of maintenance operations; The skeleton joint criticality analysis module is used to obtain the second data of each maintenance operation layer based on information entropy analysis of the maintenance demonstration operation, and the second data includes the skeleton joint criticality; The real-time adjustment module is used to obtain the multi-source video stream of the actual maintenance operation, and adjust the second data according to the misoperation actions to obtain the dynamic second data; The operation recognition module is used to recognize the actual maintenance operation based on ST-GCN according to the second skeleton feature, and the second skeleton feature is obtained by processing the acquired first skeleton feature by combining the dynamic second data.

Images