Distribution network misoperation behavior risk identification method based on graph convolutional network

Abstract

The invention discloses a distribution network misoperation behavior risk identification method based on a graph convolutional network, and relates to the field of power distribution network safety. The method comprises the following steps: receiving an operation video sequence of a distribution network operator on a distribution network site; extracting a skeleton topological graph sequence based on the operation video sequence; based on a space attention mechanism, determining an attention characteristic graph sequence according to the skeleton topological graph sequence; using the trained power distribution network operator misoperation identification model to determine an operation behavior risk identification result about the power distribution network operator according to the attention characteristic graph sequence and the skeleton topological graph sequence; wherein the power distribution network operator misoperation identification model is constructed based on a GCN model and a GRU model. Compared with the prior art, the accuracy of risk behavior identification can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network security, and more specifically, to a method for identifying risks of misoperation behaviors in a distribution network based on a graph convolutional network. Background Art

[0002] In recent years, the intelligent level of the distribution network has been continuously improved, but misoperation is still one of the important factors threatening the security of the distribution network. Due to the complex distribution network equipment, cumbersome operation steps and high precision requirements, once a mistake occurs, it is often difficult to detect and correct it in time, thus bringing major security risks. Therefore, establishing an efficient misoperation identification and prevention and control mechanism is of great significance for reducing distribution network accidents.

[0003] The traditional method relying on on-site supervision by personnel cannot achieve real-time monitoring and dynamic early warning. Especially in case of emergencies or complex operation environments, it is difficult to effectively prevent misoperations. With the progress of artificial intelligence and image recognition technologies, intelligent misoperation detection schemes have been studied at home and abroad. Image data during the operation process is collected through cameras, and deep learning models are combined to analyze operation actions to identify abnormal or misoperation behaviors in real time. However, these models lack key point (such as the hand actions of operators, equipment touch points, etc.) focus. When dealing with high frame rate and high resolution videos, the performance often drops significantly, affecting the recognition efficiency. In addition, existing misoperation identifications often do not consider the coherence and temporal relationship of actions, which are key factors for judging whether an operation is correct. Due to ignoring the time sequence between actions, the system is difficult to effectively identify continuous misoperations in complex operation scenarios. Summary of the Invention

[0004] The present invention provides a method for identifying risks of misoperation behaviors in a distribution network based on a graph convolutional network to overcome the defect of poor recognition accuracy in the above-mentioned existing technologies.

[0005] To solve the above technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a method for identifying risks of misoperation behaviors in a distribution network based on a graph convolutional network includes:

[0007] Receiving an operation video sequence of distribution network operators at the distribution network site;

[0008] Extracting a skeleton topology graph sequence based on the operation video sequence;

[0009] Determining an attention feature graph sequence based on the skeleton topology graph sequence according to a spatial attention mechanism;

[0010] Using the trained misoperation recognition model for distribution network operators, based on the sequence of attention feature maps and the sequence of skeleton topology maps, determine the recognition result of the operation behavior risk of the distribution network operators; wherein, the misoperation recognition model for distribution network operators is constructed based on the GCN model and the GRU model.

[0011] In a second aspect, a computer program product includes a computer program or computer-executable instructions, which, when executed by a processor, implement the method described in the first aspect.

[0012] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0013] The present invention discloses a method for identifying the risk of misoperation behavior in a distribution network based on a graph convolutional network. By extracting the sequence of skeleton topology maps to obtain the limb movement information of distribution network operators, through a spatial attention mechanism, the key node information in the sequence of skeleton topology maps is highlighted, and a misoperation recognition model for distribution network operators constructed based on the GCN model and the GRU model is used to capture spatial features and temporal features, thereby effectively identifying the risk behaviors of distribution network operators. Compared with the prior art, the present invention can effectively capture the dynamic process and temporal continuity of actions, and thus improve the recognition accuracy. Description of the Drawings

[0014] Figure 1 It is a schematic flow chart of a method for identifying the risk of misoperation behavior in a distribution network based on a graph convolutional network in Embodiment 1 of the present application. Detailed Embodiments

[0015] The terms "first", "second", etc. in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances, which is only a way of distinguishing objects with the same attributes when describing the embodiments of the present application. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, so that a process, method, system, product or device including a series of units does not have to be limited to those units, but may include other units not clearly listed or inherent to these processes, methods, products or devices. The term "determine" broadly covers a variety of actions, which may include obtaining, calculating, computing, processing, deriving, researching, searching (e.g., searching in a table, database or other data structure), ascertaining, and similar actions, and may also include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory) and similar actions, and may also include generating, creating, establishing and similar actions, as well as parsing, selecting, choosing and similar actions, etc. The relevant definitions of other terms will be given in the following description.

[0016] It should be noted that when a component is considered to be "connected" to another component, it can be directly connected to the other component or connected to the other component through an intermediate component. In addition, in the following embodiments, "connection", if there is transmission of electrical signals or data between the connected objects, should be understood as "electrical connection", "communication connection", etc.

[0017] The accompanying drawings are only for illustrative purposes and should not be construed as limitations on this patent;

[0018] To better illustrate this embodiment, some components in the accompanying drawings are omitted, enlarged or reduced, which does not represent the size of the actual product;

[0019] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0020] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Embodiment 1

[0022] This embodiment provides a risk identification method for misoperation behaviors of distribution networks based on graph convolutional networks. Refer to Figure 1 , including:

[0023] Receive the operation video sequence of distribution network operators at the distribution network site;

[0024] Extract the skeleton topology graph sequence based on the operation video sequence;

[0025] Based on the spatial attention mechanism, determine the attention feature map sequence according to the skeleton topology graph sequence;

[0026] Use the trained misoperation identification model for distribution network operators to determine the risk identification result of the operation behavior of the distribution network operators according to the attention feature map sequence and the skeleton topology graph sequence; wherein, the misoperation identification model for distribution network operators is constructed based on the GCN (Graph Convolutional Network) model and the GRU (Gated Recurrent Unit) model.

[0027] It should be noted that in this embodiment, the spatial attention mechanism is used to make the features of key areas such as hands and feet more prominent, so as to enhance the attention of the misoperation identification model for distribution network operators to key areas; the misoperation identification model for distribution network operators constructed based on GCN-GRU can effectively capture the spatial dependence relationship and time variation of the actions of distribution network operators, so as to effectively understand the dynamic process and temporal continuity of the actions, and further improve the accuracy of risk behavior identification.

[0028] In some preferred embodiments, determining the attention feature map sequence according to the skeleton topology map sequence includes:

[0029] Determining the key node weights for each undirected graph in the skeleton topology map sequence based on the average sternum distance, and its expression is as follows:

[0030]

[0031] In the formula, and respectively represent the distances between joint point j and the sternum node at time t and t+Δt; k represents the sternum node; represents the average distance between joint point j and the sternum within the time interval Δt;

[0032] Performing weighted and max-pooling operations on the skeleton topology map sequence according to the key node weights to determine the weighted pooling matrix X′;

[0033] Introducing a spatial attention layer, determining a dynamic attention weight matrix according to the weighted pooling matrix, and forming the attention feature map sequence, and its process is expressed as:

[0034] A s = softmax(W2·ReLU(W1·X′+b1)+b2)

[0035] In the formula, A s represents the dynamic attention weight matrix reflecting the importance of each joint point, which is a learnable matrix; W1 and W2 represent learnable weight matrices for linearly transforming node features; b1 and b2 represent the bias terms of this linear transformation; softmax(·) is used to normalize the result into a probability distribution so that the weights of each node sum to 1 in the spatial dimension, thereby more intuitively representing the importance of each node.

[0036] In the above preferred embodiments, the average sternum distance is introduced as a guiding factor to assign weights to each image region, making the recognition weights of the limbs higher. The image features are weighted according to the calculated attention weights, making the features of key regions such as hands and feet more prominent, while weakening the influence on unimportant regions.

[0037] In some alternative embodiments, determining the recognition result of the operation behavior risk of the distribution network operator includes:

[0038] Using the GCN model in the misoperation recognition model of the distribution network operator, extracting a spatial feature sequence based on the skeleton topology map sequence and the attention feature map sequence;

[0039] Using the GRU model in the misoperation identification model for distribution network operators, time features are extracted based on the spatial feature sequence to capture dependencies in time.

[0040] Using a Softmax classifier, the risk identification result of the operation behavior is output according to the time features.

[0041] It should be emphasized that using GCN can extract the spatial structure information (spatial features) of each frame of the skeleton topology graph and capture the spatial dependencies between different key nodes; while GRU can effectively learn the dynamic changes (time features) in the timing of the actions of distribution network operators through its cyclic structure. Compared with the prior art, the above embodiments can not only effectively extract the spatial dependencies of key points, but also capture the timing features of actions, so as to more accurately identify and judge continuous misoperation behaviors.

[0042] In some specific implementation processes, the risk identification result of the operation behavior includes the probability values of various operation risk behavior categories (or incorrect operation behaviors), and the risk levels that may occur when the corresponding distribution network operator continues to operate based on the probability values.

[0043] Furthermore, the expression for extracting the spatial feature sequence is as follows:

[0044]

[0045] In the formula, Ht represents the t-th frame of spatial features in the spatial feature sequence, t = {1, 2,..., T}; A represents the adjacency matrix of the skeleton topology graph, which is used to represent the connection relationship between nodes; D is the degree matrix of A, which is used to normalize the adjacency matrix; ⊙ represents element-wise multiplication, which is used to apply the weights in A s to the skeleton topology graph X t , so that GCN pays more attention to key nodes; W GCN represents the weight parameter of the GCN model.

[0046] In the above embodiments, the dynamic attention weight matrix A s is used to perform feature aggregation on the skeleton topology graph, so as to strengthen the information of the key skeleton joints of the action.

[0047] It should be noted that the spatial feature sequence is composed of a series of spatial features.

[0048] Furthermore, the expression for extracting time features based on the spatial feature sequence is as follows:

[0049] z t = σ(W z H t + U z ht-1 +b z )

[0050] r t = σ(W r H t + U r h t-1 +b r )

[0051]

[0052] wherein, H t represents the spatial feature of the t-th frame in the spatial feature sequence, t = {1, 2, …, T}; h t-1 represents the hidden state of the GRU model at the previous time step; z t represents the update gate, which is used to control the ratio of the previous state and the current state; r t represents the reset gate, which is used to determine the degree of retention of the previous state information; represents the candidate hidden state at the current time step; h t represents the current hidden state; b z , b r , b h represent the bias terms.

[0053] In some alternative embodiments, the expression for determining the weighted pooling matrix X' is as follows:

[0054]

[0055] wherein, the MaxPool(·) operation represents performing max pooling on the number of joint points, i.e., the node dimension V, to obtain an output containing only the most important nodes, namely the feature matrix X' after max pooling and weighted processing; P represents the original node feature matrix for characterizing the human joint point information. Weighted max pooling can help the model automatically focus on the action changes of the feet or hands that are far from the body, while reducing the interference of other unimportant nodes.

[0056] In some preferred embodiments, extracting the skeleton topology graph sequence based on the job video sequence includes:

[0057] Using the OpenPose state estimation algorithm, extracting human joint point information from the image frames in the job video sequence, constructing an undirected graph reflecting the human pose as the skeleton topology graph, and combining to obtain a skeleton topology graph sequence for the complete action.

[0058] It should be understood that the skeleton topology graph sequence is composed of a series of skeleton topology graphs, denoted as Where C represents the coordinate dimension, T represents the number of frames included in the skeleton topology graph sequence, and N represents the number of human joints; in the undirected graph (i.e., the skeleton topology graph), nodes represent joints such as shoulders, elbows, and hands of the human body, and edges represent joint connection relationships.

[0059] In some preferred embodiments, the training process of the misoperation recognition model for distribution network operators includes:

[0060] Obtain at least one sequence of error operation videos of distribution network operators; among them, the error operation video sequence is labeled with the category of error operation behavior.

[0061] Extract the training skeleton topology graph sequence according to the error operation video sequence.

[0062] Based on the spatial attention mechanism, determine the training attention feature map sequence according to the training skeleton topology graph sequence.

[0063] Construct the initial misoperation recognition model for distribution network operators.

[0064] Construct the training skeleton topology graph sequence, the training attention feature map sequence and the corresponding annotations into a supervised data set, and let the misoperation recognition model for distribution network operators be trained on the supervised data set and update the model parameters.

[0065] It should be understood that the video objects in the error operation video sequence are various error operation actions, including but not limited to incorrect switchgear operations, live closing, etc., and each error operation video sequence focuses on a specific error operation behavior.

[0066] In some preferred embodiments, before extracting the skeleton topology graph sequence or the training skeleton topology graph sequence, redundant frames in the operation video sequence or the error operation video sequence are removed based on similarity analysis, including:

[0067] Decode the operation video sequence or the error operation video sequence into a number of image frames.

[0068] Calculate the similarity measure between adjacent image frames:

[0069]

[0070] In the formula, F i,k represents the k-th pixel value in the pixel matrix corresponding to the i-th image frame F i ; F j,k represents the k-th pixel value in the pixel matrix corresponding to the j-th image frame F j ; S i,j represents the average pixel difference between the i-th image frame and the j-th image frame; N represents the total number of pixels.

[0071] When the average pixel difference is less than or does not exceed a preset threshold, F i and F j are regarded as similar frames, and any one of them is retained as a key frame;

[0072] The finally retained key frames are re-encoded as the job video sequence or the error job video sequence.

[0073] It should be noted that screening key frames based on image similarity can effectively remove redundant frames, obtain representative image frames, thereby effectively reducing the processing volume of redundant frames in the video, reducing the computing power requirement, and is particularly suitable for high frame rate and high resolution scenarios. For example, when monitoring the work of substation operators, only key frames are retained, thereby improving the system response speed and reducing the computing burden. At the same time, introducing a spatial attention mechanism enables the model to focus on the key parts of the wrong behaviors of distribution network operators. The above preferred embodiments significantly improve the accuracy of detecting misoperations while ensuring real-time performance, and can also perform excellently even in complex operation environments such as substations and high-voltage switch operation sites.

[0074] In some specific implementation processes, for the training skeleton topology graph sequence, after screening out key frames based on similarity, key frames that can significantly reflect the behavior changes of distribution network operators are further selected manually.

[0075] This embodiment also proposes a distribution network misoperation behavior risk identification system based on a graph convolutional network, including:

[0076] A skeleton topology graph extraction module, configured to receive a job video sequence of a distribution network operator at a distribution network site; and also configured to extract a skeleton topology graph sequence based on the job video sequence;

[0077] An attention feature map extraction module, configured to determine an attention feature map sequence based on a spatial attention mechanism according to the skeleton topology graph sequence;

[0078] A job behavior risk identification module, configured to carry a trained misoperation identification model for distribution network operators; and also configured to use the misoperation identification model for distribution network operators to determine a job behavior risk identification result of the distribution network operator according to the attention feature map sequence and the skeleton topology graph sequence; wherein, the misoperation identification model for distribution network operators is constructed based on a GCN model and a GRU model.

[0079] Embodiment 2

[0080] This embodiment provides a computer-readable storage medium, on which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, the code set or the instruction set is loaded and executed by a processor, so that the processor executes some or all of the steps of the method provided in Embodiment 1 of this application.

[0081] It can be understood that the storage medium can be transient or non-transient. Exemplarily, the storage medium includes but is not limited to various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0082] Exemplarily, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), a field programmable gate array (FPGA), or the like.

[0083] Exemplarily, the read-only memory includes but is not limited to MASK ROM, PROM, EPROM, EEPROM, Flash, etc.

[0084] Exemplarily, the random access memory includes but is not limited to DRAM, SRAM, SDRAM, DDR SDRAM, etc.

[0085] In some examples, a computer program product is provided, which can be specifically implemented in a way of hardware, software, or a combination thereof. As a non-limiting example, the computer program product can be embodied as the storage medium, and can also be embodied as a software product, such as an SDK (Software Development Kit), etc.

[0086] As a non-limiting example, a computer program product is provided. The computer program product includes a computer program or computer-executable instructions, and the computer program or computer-executable instructions are stored in a computer-readable storage medium. The processor of the electronic device reads the computer program or the computer-executable instructions from the computer-readable storage medium, and the processor executes the computer-executable instructions, so that the electronic device executes some or all of the steps of the method described in the embodiments of this application.

[0087] In some examples, a computer program is provided, including computer-readable code. When the computer-readable code runs in a computer device, a processor in the computer device executes some or all of the steps for implementing the method.

[0088] This embodiment also proposes an electronic device, including a memory and a processor. The memory stores at least one instruction, at least one program, a code set or an instruction set. When the processor executes the at least one instruction, at least one program, the code set or the instruction set, some or all of the steps of the method described in Embodiment 1 are implemented.

[0089] In some examples, a hardware entity of the electronic device is provided, including: a processor, a memory and a communication interface; wherein, the processor generally controls the overall operation of the electronic device; the communication interface is used to enable the electronic device to communicate with other terminals or servers through a network; the memory is configured to store instructions and applications executable by the processor, and can also cache data to be processed or already processed by the processor and each module in the electronic device (including but not limited to image data, audio data, voice communication data and video communication data), and can be implemented by flash memory (FLASH), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or random access memory (RAM).

[0090] The processor may include one or more processing elements. Therefore, the processor may include one or more integrated circuits (ICs) configured to execute the functions of the processor. In addition, each integrated circuit may include circuits (such as a first circuit, a second circuit, and other circuits, etc.) configured to execute the functions of the processor.

[0091] Furthermore, data transmission may be performed between the processor, the communication interface and the memory through a bus. The bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together.

[0092] It can be understood that the optional items in the above Embodiment 1 are equally applicable to this embodiment, so they will not be described repeatedly here.

[0093] The same or similar reference numerals correspond to the same or similar components;

[0094] The terms describing the positional relationship in the drawings are only for illustrative purposes and cannot be construed as a limitation to this application;

[0095] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other.

[0096] In different specific implementations, the method or system described in the present application may be implemented in software, hardware, or a combination thereof. In addition, the order of the steps of the method may be changed, and various elements may be added, reordered, combined, omitted, modified, etc.

[0097] Obviously, the above embodiments of the present application are merely examples given for clearly explaining the present application, rather than limiting the implementation manners of the present application, and are not used to limit the present application. For those of ordinary skill in the art, other different forms of changes or modifications can be made on the basis of the above description. Each discrete structural / functional module or unit may be integrated together to form an independent part, or each module may exist alone, or two or more modules may be integrated to form an independent part. The structures and functions of the discrete components can be implemented as combined structures or components. It is not necessary and impossible to enumerate all the implementation manners here. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the claims of the present application.

Claims

1. A risk identification method for misoperation behavior in a distribution network based on a graph convolutional network, characterized in that, Including: Receiving an operation video sequence of distribution network operators at the distribution network site; Extracting a skeleton topology graph sequence based on the operation video sequence; Based on the spatial attention mechanism, determining an attention feature map sequence according to the skeleton topology graph sequence; Using the trained misoperation recognition model of distribution network operators, determining the operation behavior risk recognition result of the distribution network operators according to the attention feature map sequence and the skeleton topology graph sequence; wherein, the misoperation recognition model of distribution network operators is constructed based on the GCN model and the GRU model.

2. The risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 1, characterized in that The determining the attention feature map sequence according to the skeleton topology graph sequence includes: Determining the key node weights of each undirected graph in the skeleton topology graph sequence based on the average sternum distance, and its expression is as follows: Wherein, and respectively represent the distances between the joint point j and the sternum node at times t and t+Δt; k represents the sternum node; represents the average distance between the joint point j and the sternum within the time interval Δt; Performing a weighted and max pooling operation on the skeleton topology graph sequence according to the key node weights to determine a weighted pooling matrix X'; Introducing a spatial attention layer, determining a dynamic attention weight matrix according to the weighted pooling matrix, and forming the attention feature map sequence, and its process is expressed as: A s = softmax(W2·ReLU(W1·X′ + b1) + b2) where, A s represents a dynamic attention weight matrix reflecting the importance of each joint point; W1 and W2 represent learnable weight matrices used for linear transformation of node features; b1 and b2 represent the bias terms of this linear transformation; softmax(·) is used to normalize the result into a probability distribution so that the weights of each node sum to 1 in the spatial dimension.

3. The risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 2, characterized in that, The determining the operation behavior risk recognition result of the distribution network operators includes: Using the GCN model in the misoperation recognition model of distribution network operators to extract a spatial feature sequence based on the skeleton topology graph sequence and the attention feature map sequence; Using the GRU model in the misoperation recognition model of distribution network operators to extract temporal features based on the spatial feature sequence to capture temporal dependencies; Using a Softmax classifier to output the operation behavior risk recognition result according to the temporal features.

4. The risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 3, characterized in that The expression for extracting the spatial feature sequence is as follows: where H t represents the t-th frame spatial feature in the spatial feature sequence, where t = {1, 2, …, T}; A represents the adjacency matrix of the skeleton topology graph, which is used to represent the connection relationship between nodes; D is the degree matrix of A, which is used to normalize the adjacency matrix; ⊙ represents the element-wise product, which is used to apply the weights in A s to the skeleton topology graph X t , so that the GCN pays more attention to the key nodes; W GCN represents the weight parameters of the GCN model.

5. The risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 3, characterized in that, The expression for extracting temporal features based on the spatial feature sequence is as follows: z t = σ(W z H t + U z h t-1 + b z ) r t = σ(W r H t + U r h t-1 + b r ) where, H t represents the t-th frame spatial feature in the spatial feature sequence, t = {1, 2, …, T}; h t-1 represents the hidden state of the GRU model at the previous time step; z t represents the update gate, which is used to control the proportion of the previous state and the current state; r t represents the reset gate, which is used to determine the degree of retention of the previous state information; represents the candidate hidden state at the current time step; h t represents the current hidden state; b z , b r , b h represent the bias terms.

6. The risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 2, characterized in that, The expression for determining the weighted pooling matrix X' is as follows: In the formula, the MaxPool(·) operation represents max pooling on the number of joint points, i.e., the node dimension V; P represents the original node feature matrix for characterizing the human joint point information.

7. A risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 1, characterized in that The extracting the skeleton topology graph sequence based on the operation video sequence includes: Using the OpenPose state estimation algorithm to extract human joint point information from the image frames in the operation video sequence, constructing an undirected graph reflecting the human posture as the skeleton topology graph, and combining to obtain a skeleton topology graph sequence for a complete action.

8. A risk identification method for misoperation behavior of distribution network based on graph convolutional network according to claim 1, characterized in that The training process of the misoperation recognition model of distribution network operators includes: Obtaining at least one wrong operation video sequence of distribution network operators; wherein, the wrong operation video sequence is labeled with the wrong operation behavior category; Extracting a training skeleton topology graph sequence according to the wrong operation video sequence; Based on the spatial attention mechanism, determining a training attention feature map sequence according to the training skeleton topology graph sequence; Constructing the initial misoperation recognition model of the distribution network operators; Constructing the training skeleton topology graph sequence, the training attention feature map sequence and the corresponding annotations into a supervised data set, and making the misoperation recognition model of the distribution network operators train on the supervised data set and update the model parameters.

9. A risk identification method for misoperation behavior of a distribution network based on a graph convolutional network according to any one of claims 1-8, characterized in that, Before extracting the skeleton topology graph sequence or the training skeleton topology graph sequence, redundant frames in the operation video sequence or the incorrect operation video sequence are removed based on similarity analysis, including: Decoding the operation video sequence or the incorrect operation video sequence into a number of image frames; Calculating the similarity metric between adjacent image frames; Where F i,k represents the k-th pixel value in the pixel matrix corresponding to the i-th image frame F i ; F j,k represents the k-th pixel value in the pixel matrix corresponding to the j-th image frame F j ; S i,j represents the average pixel difference between the i-th image frame and the j-th image frame; N represents the total number of pixels. When the average pixel difference is less than or does not exceed a preset threshold, F i and F j are regarded as similar frames, and any one of them is retained as a key frame; Recoding the finally retained key frames as the operation video sequence or the incorrect operation video sequence.

10. A computer program product, comprising a computer program or computer-executable instructions, characterized in that, When the computer program or computer-executable instructions are executed by a processor, the method according to any one of claims 1-9 is implemented.

Images