Protection system for live working of power distribution network

Abstract

The invention discloses a protection system for live-line work of a power distribution network, and the system comprises the steps: shooting a large number of training images from a current common live-line work training system, and carrying out the edge repairing of the large number of shot training images through employing a filtering algorithm, and obtaining a repaired image; inputting the repaired image into a behavior recognition model for training to obtain a trained behavior recognition model; and shooting actual hot-line work to obtain real-time image data, inputting the real-time image data into the trained behavior recognition model for processing to obtain a final semantic segmentation mask, and judging the hot-line work of a worker according to the final semantic segmentation mask. According to the invention, through the behavior identification model, the identification effect is optimized, the accuracy in practical application is improved, and guiding or early warning is carried out on the working personnel in operation in real time.

Description

Technical Field

[0001] This invention relates to the field of live-line working technology in power distribution networks, specifically to a protection system for live-line working in power distribution networks. Background Technology

[0002] Live-line working in power distribution networks refers to maintenance and repair work performed on power distribution networks (typically at voltage levels of 10kV and below, responsible for power distribution) without interrupting power supply. Unlike ultra-high voltage transmission line work (which involves equipotential bonding and wearing protective clothing), live-line working, due to the dense equipment and confined space, primarily employs indirect methods such as insulation wrapping or isolation, demanding extremely high levels of skill and psychological fortitude. Live-line working has become a key technology for ensuring power supply reliability.

[0003] The core challenges of live-line work in power distribution networks include: 1) High operational risks: The small spacing between power distribution lines and the complexity of some lines increase the risk of electric shock, single-phase grounding, or phase-to-phase short circuits during long-term operations, potentially leading to personal injury or system failure; 2) Outdoor work may result in short circuits or leakage due to high humidity or dust accumulation. Severe weather conditions such as wind and rain further exacerbate these risks. Current solutions to these problems primarily include: conducting non-live-line simulation training to improve basic skills through insulated tool operation training; using personal protective equipment such as insulated clothing, gloves, and shoes to provide basic insulation protection; employing specialized protection devices for distribution boxes, such as dehumidification and temperature control devices, to reduce the risk of environmentally-related electrical faults; and minimizing work in windy or rainy weather.

[0004] However, the above solutions still have significant limitations: simulated training is actually a non-energized scenario, which cannot reproduce the real electric field environment, and trainees' psychological adaptability is not developed, resulting in a gap between training outcomes and actual needs. Live-line training, on the other hand, carries a high risk of electric shock due to trainees' lack of skills, and there is a lack of safe and controllable transitional training methods. Personal protective equipment has limited effectiveness, and the protective effect of insulating equipment is limited by material aging, specification matching, and working posture. Moreover, using a large number of complex and redundant equipment is not conducive to the implementation of operations. Simply stating that one should not work or should work with caution in windy and rainy weather is not very effective in practice. How to judge the intensity of wind and rain and how to quantify the level of environmental hazard are not introduced, and in some cases, emergency repairs need to be carried out in severe weather. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a protection system for live-line work in power distribution networks, solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a protection system for live-line working in distribution networks, comprising: The data acquisition module is used to capture a large number of training images from commonly used live-line working training systems. The captured training images are processed by using a filtering algorithm to repair edges and obtain repaired images. The training module is used to input the repaired image into the behavior recognition model for training, and obtain the trained behavior recognition model. The feature processing module is used to capture images of actual live-line work, acquire real-time image data, input the real-time image data into the trained behavior recognition model for processing, and obtain the final semantic segmentation mask. The prediction and judgment module is used to judge the live-line work of the staff based on the final semantic segmentation mask.

[0007] Furthermore, the behavior recognition model consists of an encoder, a Transformer Block module, and a decoder; The encoder consists of a spatial and channel collaborative attention module and a cross-layer feature pyramid module; The Transformer Block module consists of a first channel, a second channel, a third channel, and a fourth channel; The decoder consists of multiple lightweight MLP layers, dilated spatial convolutional pooling pyramid modules, and classification and segmentation convolutions; Real-time image data is sequentially input into the spatial and channel collaborative attention module and the cross-layer feature pyramid module to perform block, step size, and fill overlapping image embedding operations, and the output of the cross-layer feature pyramid module is obtained. The output of the cross-layer feature pyramid module is input into the first channel, the second channel, the third channel, and the fourth channel for processing to obtain the first multi-scale feature map, the second multi-scale feature map, the third multi-scale feature map, and the fourth multi-scale feature map. The first, second, third, and fourth multi-scale feature maps are sequentially input into multiple lightweight MLP layers, dilated spatial convolution pooling pyramid modules, and classification and segmentation convolutions for processing, and the final semantic segmentation mask is output.

[0008] Furthermore, the specific process for obtaining the repaired image is as follows: Perform supergreen segmentation on the training image to obtain the corresponding supergreen segmented image. Iterate through each non-zero pixel in the supergreen segmented image and use the Sobel operator's convolution kernel to calculate the gradient magnitude of the brightness change rate in the four horizontal directions (i-1,j)(i+1,j) and vertical directions (i,j-1)(i,j+1) for each non-zero pixel. , , , , i.e., the horizontal derivative and the vertical derivative; i, j are the i-coordinates and j-coordinates of the pixel in the super-green segmentation image, respectively; This represents the magnitude of the brightness gradient in the westward direction of the pixel. This represents the magnitude of the brightness gradient in the east direction of the pixel. This represents the magnitude of the brightness gradient in the north direction of the pixel. This represents the magnitude of the brightness gradient in the south direction of the pixel. The Perona-Malik model is used to control the anisotropy of the horizontal and vertical derivatives, determining the diffusion coefficients in the four directions. Based on these coefficients, the non-zero pixel at position (i,j) in the supergreen segmentation image is calculated, thus obtaining the value of any pixel. The gradient ∇·I in the four directions of north, south, east, and west; Let be the brightness value of the non-zero pixel at position (i,j) in the supergreen segmentation image traversed in the previous text, at the t-th iteration. And apply the edge stopping function g(|∇·I|) to any pixel point. The gradient ∇·I in the four directions of north, south, east, and west is weighted; the weighted arbitrary pixel point The gradient ∇·I in the four directions of north, south, east, and west is multiplied by the coefficient λ and then summed to obtain the update amount of the current pixel. The update amount of the current pixel is then used to calculate... The values ​​are then iterated over until a stopping condition is met. Gradient energy threshold Set to 0.1, when less than Output This is the repaired image; For the non-zero pixel at position (i,j) in the supergreen segmentation image traversed previously, in the i-th... The brightness value during the round of iteration.

[0009] Furthermore, the specific process of training the behavior recognition model is as follows: During the training of the behavior recognition model, the repaired images are labeled to indicate the categories of human bodies, tools, and charged objects in the repaired images, which serve as the ground truth labeling mask; the repaired images with completed labeling are divided into training sets and validation sets; The repaired images in the training set are input into the behavior recognition model in batches for processing, and the behavior recognition model predicts and generates the semantic segmentation mask of the training set. Calculate the weighted cross-entropy loss, Dice loss, and safety distance perception loss designed for the scene by comparing the semantic segmentation mask of the training set with the real labeled mask of the training set. The weighted cross-entropy loss, Dice loss, and safe distance perception loss are summed to form a composite loss function. The gradients of the parameters of each layer of the behavior recognition model are calculated based on the composite loss function; the parameters of the behavior recognition model are updated using the AdamW optimizer, and the learning rate of the behavior recognition model is scheduled using the cosine annealing strategy. The training is completed when the parameter update magnitude reaches the preset threshold, the learning rate decays to the set target value, and the performance index of the behavior recognition model on the validation set converges stably. The validation set is input into the behavior recognition model to predict the semantic segmentation mask of the validation set. Based on the predicted semantic segmentation mask of the validation set and the true labeled mask of the validation set, the composite loss and IoU performance index corresponding to the semantic segmentation mask of the validation set are calculated. When the fluctuation range of the composite loss and IoU performance index is continuously lower than the set threshold, it is determined that the performance index of the behavior recognition model on the validation set has reached stable convergence.

[0010] Furthermore, the specific process of judging the worker's live-line work based on the final semantic segmentation mask is as follows: The judgment of live-line work includes the judgment of safe distance intrusion, the judgment of missing personal protective equipment, and the judgment of omission of key steps; Intrusion detection at safe distance: Extract pixel clusters of human body / tool ​​and charged object from the final semantic segmentation mask; calculate the minimum Euclidean distance between the two pixel clusters of human body / tool ​​and charged object in the repaired image; map the distance to the real three-dimensional space according to the camera calibration parameters and the known size of the charged object device; and determine whether the set safe distance is met. Personal protective equipment (PPE) shortage judgment: Locate a specific area in the final semantic segmentation mask, and check whether the confidence of the insulating glove category in the located specific area is lower than the threshold; at the same time, check whether there is a high confidence prediction of the insulating clothing category in the human torso area. If any condition is not met, a PPE shortage warning is triggered. Key step omission detection: Establish a temporal buffer of length N; extract N frames of key target states from the final semantic segmentation mask and store the extracted N frames of key target states into the temporal buffer; Define corresponding verification rules based on the actual operation specifications for live-line work in distribution networks; Based on the defined verification rules, the live-line operation of the power distribution network captured in real time is identified, and the preceding steps corresponding to the operation to be performed are determined. Check the time-series cache to see if the preceding steps corresponding to the current operation have been completed; if the completion status of the preceding steps corresponding to the current operation is not detected in the time-series cache, trigger a step omission warning.

[0011] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a protection system for live-line work in a power distribution network.

[0012] A non-volatile computer storage medium storing computer-executable instructions, the computer being able to execute a protection system for live-line work in a power distribution network.

[0013] Compared with existing technologies, the present invention has the following advantages: This invention overcomes the shortcomings of conventional models, which cannot be trained due to the lack of a large amount of actual live-line operation image data. At the same time, it uses a behavior recognition model to optimize the recognition effect, thereby improving the accuracy in practical applications and providing real-time guidance or warnings to workers who are working. Attached Figure Description

[0014] Figure 1 This is a system flowchart of the present invention.

[0015] Figure 2 This is a schematic diagram of the behavior recognition model structure of the present invention. Detailed Implementation

[0016] like Figure 1 As shown, the present invention provides a technical solution: a protection system for live-line working in a power distribution network, comprising: The data acquisition module is used to capture a large number of training images from commonly used live-line working training systems. The captured training images are processed by using a filtering algorithm to repair edges and obtain repaired images. The training module is used to input the repaired image into the behavior recognition model for training, and obtain the trained behavior recognition model. The feature processing module is used to capture images of actual live-line work, acquire real-time image data, input the real-time image data into the trained behavior recognition model for processing, and obtain the final semantic segmentation mask. The prediction and judgment module is used to judge the live-line work of the staff based on the final semantic segmentation mask.

[0017] like Figure 2 As shown, the behavior recognition model consists of an encoder, a Transformer Block module, and a decoder; The encoder consists of a Spatial and Channel Cooperative Attention Module (SCSA) and a Cross-Layer Feature Pyramid (CFPT) module; The Transformer Block module consists of a first channel, a second channel, a third channel, and a fourth channel; The decoder consists of multiple lightweight MLP layers, a dilated spatial convolutional pooling pyramid (ASPP) module, and a classification and segmentation convolution (Class Seg); Real-time image data is sequentially input into the spatial and channel collaborative attention module and the cross-layer feature pyramid module to perform block, step size, and fill overlapping image embedding operations, and the output of the cross-layer feature pyramid module is obtained. The output of the cross-layer feature pyramid module is input into the first channel, the second channel, the third channel, and the fourth channel for processing, respectively, to obtain the first multi-scale feature map, the second multi-scale feature map, the third multi-scale feature map, and the fourth multi-scale feature map. The first, second, third, and fourth multi-scale feature maps are sequentially input into multiple lightweight MLP layers, dilated spatial convolution pooling pyramid modules, and classification and segmentation convolutions for processing, and the final semantic segmentation mask is output.

[0018] The specific process for obtaining the repaired image is as follows: For the edges of human bodies or devices in the training images, filtering algorithms can be used to repair the edges, specifically: First, perform super-green segmentation on the training image to obtain the corresponding super-green segmented image. Then, traverse each non-zero pixel in the super-green segmented image (taking the non-zero pixel at position (i,j) as an example), and use the convolution kernel of the Sobel operator to calculate the gradient magnitude of the brightness change rate in the four horizontal directions (i-1,j)(i+1,j) and vertical directions (i,j-1)(i,j+1) of the non-zero pixel. , , , , i.e., the horizontal derivative and the vertical derivative; i, j are the i-coordinates and j-coordinates of the pixel in the super-green segmentation image, respectively; This represents the magnitude of the brightness gradient in the westward direction of the pixel. This represents the magnitude of the brightness gradient in the east direction of the pixel. This represents the magnitude of the brightness gradient in the north direction of the pixel. This represents the magnitude of the brightness gradient in the south direction of the pixel. The Perona-Malik model is used to anisotropically control the horizontal and vertical derivatives to determine the diffusion coefficients in the four directions. Based on the determined diffusion coefficients in the four directions, the non-zero pixel at position (i,j) in the supergreen segmentation image is calculated to obtain the arbitrary pixel. The gradient ∇·I in the four directions of North (N), South (S), East (E), and West (W); Let be the brightness value of the non-zero pixel at position (i,j) in the supergreen segmentation image traversed in the previous text, at the t-th iteration. And apply the edge stopping function g(|∇·I|) to any pixel point. The gradients ∇·I in the four directions of North (N), South (S), East (E), and West (W) are weighted; the weighted arbitrary pixel points are then... The gradient ∇·I in the four directions of North (N), South (S), East (E), and West (W) is multiplied by the coefficient λ and then summed to obtain the update amount of the current pixel. The update amount of the current pixel is then used to calculate... The values ​​are then iterated over until a stopping condition is met. Here, the gradient energy threshold Set to 0.1, when less than Output This is the repaired image; For the non-zero pixel at position (i,j) in the supergreen segmentation image traversed previously, in the i-th... The brightness value during the round of iteration.

[0019] The specific process of training the behavior recognition model is as follows: The repaired image is labeled to indicate the categories of human bodies, tools, and charged objects in the repaired image, which are used as the ground truth label mask; the repaired image with completed labeling is divided into training set and validation set; The repaired images in the training set are input into the behavior recognition model in batches for processing, and the behavior recognition model predicts and generates the semantic segmentation mask of the training set. Calculate the weighted cross-entropy loss, Dice loss, and safety distance perception loss designed for the scene by comparing the semantic segmentation mask of the training set with the real labeled mask of the training set. The weighted cross-entropy loss, Dice loss, and safe distance perception loss are summed to form a composite loss function. The gradients of the parameters of each layer of the behavior recognition model are calculated based on the composite loss function; the parameters of the behavior recognition model are updated using the AdamW optimizer, and the learning rate of the behavior recognition model is scheduled using the cosine annealing strategy. The training is completed when the parameter update magnitude reaches the preset threshold, the learning rate decays to the set target value, and the performance index of the behavior recognition model on the validation set converges stably. The validation set is input into the behavior recognition model to predict the semantic segmentation mask of the validation set. Based on the predicted semantic segmentation mask of the validation set and the true labeled mask of the validation set, the composite loss and IoU performance index corresponding to the semantic segmentation mask of the validation set are calculated. When the fluctuation range of the composite loss and IoU performance index is continuously lower than the set threshold, it is determined that the performance index of the behavior recognition model on the validation set has reached stable convergence.

[0020] Among them, the safety distance perception loss term in the composite loss function does not directly calculate the physical distance. Instead, it implicitly guides the behavior recognition model to focus on the safety distance by penalizing inappropriate closeness between the "human body / tool" category and categories such as "charged body" in the feature space or the repaired image space. Specifically: Define the centroid coordinates of the two predicted regions for these two categories. Use a composite loss function to make the distance between the centroids of the two regions greater than a preset "penalty threshold" (in pixels, which can be calculated and set according to image resolution and actual safe distance). This will force the behavior recognition model to actively "push away" these two types of objects (human body / tool, charged body) in the image when learning segmentation, so as to incorporate the prior knowledge of "safe distance" into the feature encoding stage, and further ensure that the distance between specific parts of the human body and the target electrical equipment is controllable during operation.

[0021] The specific process of judging the worker's live-line work based on the final semantic segmentation mask is as follows: The judgment of live-line work includes the judgment of safe distance intrusion, the judgment of missing personal protective equipment, and the judgment of omission of key steps; Intrusion detection based on safe distance: Extract pixel clusters for "human body / tool" and "charged object" from the final semantic segmentation mask. Calculate the minimum Euclidean distance between the two pixel clusters in the repaired image. Based on camera calibration parameters and known device dimensions, map this distance to the actual 3D space to determine if it meets the safe distance requirements specified in the database. Personal protective equipment (PPE) missing judgment: Locate a specific region (e.g., "hands") in the final semantic segmentation mask, and check whether the confidence of the "insulating gloves" category in the located specific region is lower than the threshold; at the same time, check whether there is a high confidence prediction of the "insulating clothing" category in the "human body" torso region. If any condition is not met, a "PPE missing" warning is triggered. Key step omission detection: Establish a temporal buffer of length N; extract N frames of key target states from the final semantic segmentation mask and store the extracted N frames of key target states into the temporal buffer; Define corresponding verification rules based on the actual operation specifications for live-line work in distribution networks; (for example: "Before performing the 'contact B-phase conductor' operation, the prerequisite step of 'installing the B-phase insulation shield' must be completed"). Based on the defined verification rules, the live-line operation of the power distribution network captured in real time is identified, and the prerequisite steps corresponding to the operation to be performed are determined (for example, if the B-phase conductor needs to be touched, the prerequisite step is "installing the B-phase insulation shield"). Check the time-series cache to see if the "preceding steps corresponding to the current operation" have been completed; if the completion status of the preceding steps corresponding to the current operation is not detected in the time-series cache, trigger the "step omission" warning.

[0022] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A protection system for live working on an electrical distribution network, characterized in that, The method comprises the following steps: A data acquisition module is used to capture a large number of training images from the currently commonly used live working training system, and a filtering algorithm is used to repair the edges of the large number of training images to obtain repaired images; A training module is used to input the repaired images into a behavior recognition model for training to obtain a trained behavior recognition model; A feature processing module is used to capture actual live working to obtain real-time image data, input the real-time image data into the trained behavior recognition model for processing, and obtain a final semantic segmentation mask; A prediction and determination module is used to determine the live working of the worker according to the final semantic segmentation mask.

2. The protection system for live working on power distribution networks according to claim 1, characterized in that: The behavior recognition model comprises an encoder, a Transformer Block module and a decoder; The encoder comprises a spatial and channel collaborative attention module and a cross-layer feature pyramid module; The Transformer Block module comprises a first channel, a second channel, a third channel and a fourth channel; The decoder comprises a plurality of lightweight MLP layers, a hollow spatial convolutional pooling pyramid module and a classification and segmentation convolution; The real-time image data is sequentially input into the spatial and channel collaborative attention module, the cross-layer feature pyramid module, and the execution block, step and padding overlap image embedding operation is performed to obtain the output of the cross-layer feature pyramid module; The output of the cross-layer feature pyramid module is input into the first channel, the second channel, the third channel and the fourth channel for processing to obtain a first multi-scale feature map, a second multi-scale feature map, a third multi-scale feature map and a fourth multi-scale feature map, respectively; The first multi-scale feature map, the second multi-scale feature map, the third multi-scale feature map and the fourth multi-scale feature map are sequentially input into the plurality of lightweight MLP layers, the hollow spatial convolutional pooling pyramid module and the classification and segmentation convolution for processing to output the final semantic segmentation mask.

3. The protection system for live working on power distribution networks according to claim 2, characterized in that: The specific process of obtaining the repaired images is as follows: A super green segmentation operation is performed on the training image to obtain a corresponding super green segmentation image, each non-zero pixel point of the super green segmentation image is traversed, and a convolution kernel of a Sobel operator is used to calculate gradient amplitudes of luminance change rates in four directions, i.e., horizontal left and right (i-1, j) and vertical up and down (i, j-1) and (i, j+1) of the non-zero pixel point, i.e., horizontal direction derivative and vertical direction derivative; i and j are i coordinate and j coordinate of the pixel point in the super green segmentation image 、 、 、 i and j are i coordinate and j coordinate of the pixel point in the super green segmentation image is a luminance change gradient amplitude in the west direction of the pixel point; is a luminance change gradient amplitude in the east direction of the pixel point; is a luminance change gradient amplitude in the north direction of the pixel point; is a luminance change gradient amplitude in the south direction of the pixel point; Adopt Perona-Malik model to control anisotropy of horizontal direction derivative and vertical direction derivative, determine four direction diffusion coefficients, based on the determined four direction diffusion coefficients, calculate non-zero pixel point of (i,j) position in the foregoing super green segmentation image, obtain any pixel point Gradient ∇·I in north, south, east and west four directions; Luminance value of non-zero pixel of (i,j) position in the foregoing traversed super green segmentation image at the tth iteration. and the edge stop function g(|∇·I |) is used to any pixel point The gradients ∇·I in the north, south, east and west directions are weighted; the weighted any pixel point The gradients ∇·I in the north, south, east and west directions are multiplied by the coefficient λ and then accumulated to obtain the update amount of the current pixel, and the value of the current pixel is calculated and iteration is performed until the stop condition is met , the gradient energy threshold is set to 0.1, and when it is less than output , that is, the repaired image; is the luminance value of the non-zero pixel at the (i, j) position in the super green segmentation image traversed in the foregoing text at the first iteration.

4. The protection system for live working on power distribution networks according to claim 3, characterized in that: The specific process of training the behavior recognition model is as follows: During the training of the behavior recognition model, the repaired images are labeled to mark the human body, tools and live body categories in the repaired images as real annotation masks; The labeled repaired images are divided into a training set and a verification set; The repaired images in the training set are batched and input into the behavior recognition model for processing to generate semantic segmentation masks of the training set by the behavior recognition model; The weighted cross-entropy loss value, the Dice loss value and the safety distance perception loss value designed for the scene of the semantic segmentation masks of the training set and the real annotation masks of the training set are calculated; The weighted cross-entropy loss value, the Dice loss value and the safety distance perception loss value are weighted and summed to form a composite loss function; The gradients of the parameters of each layer of the behavior recognition model are calculated based on the composite loss function; the parameters of the behavior recognition model are updated by using the AdamW optimizer, and the learning rate of the behavior recognition model is scheduled by using the cosine annealing strategy; when the parameter update amplitude reaches a preset threshold, the learning rate decays to a set target value, and the performance indicators of the behavior recognition model on the verification set are stably converged, the training is completed. The verification set is input into the behavior recognition model to predict a semantic segmentation mask of the verification set, and a composite loss corresponding to the semantic segmentation mask of the verification set, an IoU performance index are calculated based on the predicted semantic segmentation mask of the verification set and a true annotation mask of the verification set; when the fluctuation range of the composite loss and the IoU performance index is continuously lower than a set threshold, it is determined that the performance index of the behavior recognition model on the verification set is stably converged.

5. The protection system for live working on power distribution networks according to claim 4, characterized in that: The specific process of judging the live working of the worker according to the final semantic segmentation mask is as follows: The live working judgment includes safety distance invasion judgment, personal protective equipment absence judgment and key step omission judgment. The safety distance invasion judgment is as follows: the pixel clusters of the human / tool and the live body are extracted from the final semantic segmentation mask; the minimum Euclidean distance between the two pixel clusters of the human / tool and the live body in the repaired image is calculated, and the distance in the real three-dimensional space is mapped according to the camera calibration parameters and the known size of the live body equipment to judge whether the set safety distance is met. The personal protective equipment absence judgment is as follows: a specific region is located in the final semantic segmentation mask, and it is checked whether the confidence of the insulating glove category in the specific region is lower than a threshold value; at the same time, it is checked whether there is a high confidence prediction of the insulating clothing category in the human torso region, and if any condition is not met, a PPE absence warning is triggered. The key step omission judgment is as follows: a time sequence cache with a length of N is established. N frames of key target states are extracted from the final semantic segmentation mask and stored in the time sequence cache. According to the actual operation specification of the live working of the distribution network, corresponding verification rules are defined. Based on the defined corresponding verification rules, the live working operation of the distribution network actually shot is recognized to determine the corresponding pre-step of the operation to be performed. It is checked whether the pre-step corresponding to the current operation has been completed in the time sequence cache. When the completion state of the pre-step corresponding to the current operation is not detected in the time sequence cache, a step omission warning is triggered.

6. An electronic device, comprising: The system comprises a processor, a memory and a bus, the processor and the memory are connected through the bus, wherein the memory is used for storing a group of program codes, and the processor is used for calling the program codes stored in the memory to execute the protection system for live working of the distribution network according to any one of claims 1-5.

7. A non-transitory computer storage medium storing computer-executable instructions, the computer-executable instructions comprising instructions for: The computer executable instructions execute the protection system for live working of the distribution network according to any one of claims 1-5.

Images