Power grid information operation violation risk detection management and control method and device

Abstract

The invention discloses a power grid information operation violation risk detection management and control method and device, and relates to the technical field of power grid information operation. The method comprises an inspection program and a repair program, and the inspection program comprises the following steps that an operator inspects power grid equipment in a range, a recorder obtains video information during inspection in real time, a processing unit and a first shadow model cooperate to execute a video feature extraction program to obtain feature data, and the feature data is sent to the first shadow model; the feature data comprises operation actions and operation environments, the first shadow model constructs an inspection process model, and the inspection process model continuously obtains the feature data for automatic updating. By constructing the inspection process model, the inspection sequence of the power grid equipment can be automatically planned, and potential safety risks of the equipment are avoided; by constructing the maintenance tree model, the operation steps of the operator can be predicted in advance and reminded, and safety accidents caused by misoperation or out-of-range operation are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid information operation, and in particular to a method and device for detecting and controlling risk of illegal operation of power grid information. Background Art

[0002] Traditional power grid risk prevention and control technologies perform compliance checks through preset server instructions, and are suitable for out-of-scope operation identification and risk prevention and control in operations such as maintenance, inspection, and patrol in scenarios such as substations and transmission lines.

[0003] However, when equipment models are updated or operating procedures are changed, the rules need to be manually reconfigured, which is inefficient and cannot adapt to dynamic operating scenarios and new instructions. There is a lack of corresponding emergency measures for emergencies, and the real-time performance is insufficient. There is no distinction between different types of operations, which can easily lead to out-of-scope operations. At the same time, the overall risk management relies on manual post-analysis, and it is difficult to issue risk warnings before or during the execution of operating instructions, causing unnecessary losses to power grid inspections and maintenance. Therefore, improving the real-time performance and autonomous optimization of the power grid information operation risk management model is a technical problem that technical personnel in this field need to solve. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method and device for detecting and controlling risks of power grid information operations violations, which solves the problems raised in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for detecting and controlling the risk of illegal operations of power grid information, which is implemented based on a cloud service cluster, a transfer base station and an operation terminal. The cloud service cluster includes a risk assessment model, a prediction model, a first shadow model, a second shadow model and a repair background. The transfer base station includes a processing unit and a transmission unit. The operation terminal includes a recorder, a feedback module, a communication module and a speaker. The control method includes an inspection program and a repair program. The inspection program includes the following steps:

[0006] The operator inspects the power grid equipment within the signal coverage of the transfer base station. The recorder obtains the video information of the inspection in real time. The video information is transmitted to the processing unit through the communication module and the transmission unit in turn. The processing unit and the first shadow model cooperate to execute the video feature extraction program to obtain feature data. The feature data includes the operator's operation action and the operator's working environment. The first shadow model constructs the inspection process model based on the feature data. The inspection process model continuously obtains new feature data for automatic iteration and update.

[0007] When the operator finds a fault or safety defect in the power grid equipment during the inspection task, the feedback module is triggered actively. The feedback signal of the feedback module is transmitted to the nearest transfer base station processing unit through the communication module and the transmission unit in turn. The processing unit generates a repair work order and transmits it to the repair backend in the cloud service cluster through the transmission unit. After the operator completes the repair procedure, the next step is executed;

[0008] The risk assessment model presets the aging coefficient of the equipment corresponding to each repair work order , based on the aging factor Conduct risk assessment on each device, specifically adjust the inspection weight of each device , the calculation formula is , γ is the number of devices of this model within the signal coverage of the transfer base station, and the risk assessment model is based on the inspection weight of each device in the inspection process model Update the inspection order of the equipment. The inspection order is based on the inspection weight. Arranged from large to small, inspection power is great Prioritize the inspection of equipment. By continuously optimizing the inspection sequence of equipment, equipment that may pose safety risks in the power grid can be detected in advance, making it easier for operators to report maintenance orders in a timely manner.

[0009] The repair process includes the following steps:

[0010] The repair backend collects repair work orders and sends server instructions to the processing unit of the corresponding transfer base station. The processing unit sends the maintenance task corresponding to the server instruction to the corresponding operation terminal through the transmission unit. The operator corresponds to the operation terminal one by one, and the operator is assigned to the location corresponding to the maintenance task to perform the operation;

[0011] When the operator performs maintenance work, the processing unit and the second shadow model work together to execute the video feature extraction program to obtain feature data, and the second shadow model builds a maintenance tree model based on the feature data. The maintenance tree model continuously obtains new feature data for automatic iterative update;

[0012] The prediction model predicts the operator's work steps in advance according to the maintenance tree model. Before the operator's current work is completed, the prediction model sends the next step of the work prompt. Specifically, the processing unit transmits the operator's characteristic data to the prediction model. The prediction model sends the server instruction to the processing unit according to the maintenance tree model. The processing unit transmits the server instruction to the speaker. The speaker converts the server instruction into audio playback. The prediction model has built-in power safety operation procedures of the power grid company. Combined with the maintenance tree model, it predicts the risks that may be caused by the operator's work one step in advance, and sends the safety steps that need attention to the corresponding transfer base station processing unit in the form of server instructions. The processing unit converts the server instruction into a prompt voice and forwards it to the operation terminal through the transmission unit. The speaker of the operation terminal broadcasts it, which can indirectly improve the safety awareness of the on-site operators. At the same time, the above steps can also identify the on-site operators who have performed out-of-scope operations, that is, the operation steps that exceed the power grid company's power safety operation procedures. The speaker of the operation terminal also broadcasts a warning voice to remind the on-site operators to pay attention.

[0013] It should be noted that if the operation terminal directly communicates data with the server cluster, and the operation terminal is mainly for the video information obtained by the recorder, on the one hand, it will increase the computing pressure and communication bandwidth pressure of the server cluster, and there may be delays in the issuance of server instructions. On the other hand, excessive data transmission is likely to lead to unstable data transmission, while increasing the battery life and heat generation of the operation terminal. Therefore, using multiple relay base stations for signal coverage can achieve short-distance real-time data communication. The processing unit filters the video information to extract feature data to alleviate the pressure of data communication. At the same time, using multiple processing units for distributed computing can reduce the computing pressure of the server cluster, allowing the server cluster to focus on inspection program optimization and maintenance step prediction. Short-distance communication between the relay base station and the operation terminal can improve response speed, improve the real-time nature of risk detection, and avoid operators waiting on site. The prediction model sends server instructions at least 2-3 steps in advance. After receiving the processor instruction, the processing unit can send the next step to the operation terminal according to the received feedback signal, avoiding the delay problem in the issuance of server instructions.

[0014] Further, the video feature extraction procedure includes the following steps:

[0015] The processing unit fuses multiple video information to obtain a panoramic video, and the processing unit extracts operation features from the panoramic video;

[0016] Human action recognition: The processing unit uses the OpenPose algorithm detection library to extract the joint nodes of the person from the panoramic video. If the algorithm detection library does not extract the joint nodes of the operator from the panoramic video, it jumps to the working environment recognition step. Otherwise, it proceeds to the next step.

[0017] The joint nodes include the head, torso, left upper arm, left lower arm, left hand, right upper arm, right lower arm, right hand, left thigh, left calf, left foot, right thigh, right calf and right foot. The processing unit uses line segments to connect the joint nodes of the character to obtain a simplified skeleton. The processing unit extracts the simplified skeleton of each frame in the panoramic video and arranges and plays them in the order of the previous and next frames to obtain a simplified action. The processing unit transmits the simplified action to the first shadow model in the cloud service cluster through the transmission unit. The first shadow model uses the YOLOv8 image recognition algorithm to identify the operation action corresponding to the simplified action. The processing unit uses the simplified skeleton as the center line and extends a preset width of 50 pixels to both sides of the center line to obtain a widened rectangle. The processing unit removes the widened rectangle from the panoramic video and proceeds to the next step.

[0018] Working environment recognition: the processing unit uses a frame difference method to identify the picture change area in the panoramic video. The processing unit transmits the picture change area to the first shadow model through the transmission unit. The first shadow model integrates the training data set. The first shadow model is trained based on the YOLOv8 image recognition algorithm using the training data set. The first shadow model uses the image recognition algorithm to identify the working environment corresponding to the picture change area.

[0019] It should be noted that the training data sets are the irregular behavior detection data set of substation operators, the transmission line insulator defect data set, the insulating oil leakage detection image data set inside the power equipment, the GIS partial discharge defect detection data set, the distribution room status detection data set, the transmission line foreign body data set and the distribution power component defect data set. The above training data sets all use VOC labels and can be directly used for the training of the YOLOv8 image recognition algorithm.

[0020] Furthermore, the method for obtaining the panoramic video specifically includes the following steps:

[0021] Time axis synchronization: when the recorder starts recording, the recording time will be displayed in the upper right corner of the video information screen. The processing unit will synchronize the time axes of multiple video information based on the recording time and enter the next processing step;

[0022] Screen alignment: the processing unit converts the first frame of the video information into a first grayscale image, the grayscale value range of the first grayscale image is 0-255, the processing unit calculates the grayscale value difference between the pixels in the first grayscale image, normalizes the grayscale value difference of all pixels, assigns the maximum grayscale value difference to 1, assigns the minimum grayscale value difference to 0, and converts the remaining grayscale value differences in geometric proportion to the maximum and minimum values, marks the pixels whose grayscale value difference is greater than the judgment threshold as contour lines, and aligns the contour lines in all video information to obtain a fused video, and proceeds to the next processing step;

[0023] De-noising, the processing unit uses Gaussian filtering to de-noise the fused video and proceeds to the next processing step;

[0024] Distortion correction,The processing unit uses the FishEye camera model to perform distortion correction on the fused video to obtain a panoramic video.

[0025] Furthermore, the specific steps of the frame difference method for identifying the image change area in the panoramic video are as follows:

[0026] The processing unit converts each frame of the panoramic video into a second grayscale image. The grayscale value range of the second grayscale image is 0-127. The identification of the image change area only requires simple identification, and there is no need to refine the grayscale value range. The grayscale value range of 0-127 can meet the identification requirements, which can reduce a certain amount of calculation and speed up the response speed of the processing unit. The processing unit calculates the grayscale value difference between pixels in the second grayscale image, and normalizes the grayscale value difference values ​​of all pixels. The normalization process is consistent with the normalization process of the first grayscale image.

[0027] The processing unit sets a change threshold, and marks pixels in the second grayscale image whose adjacent pixels are greater than the change threshold as change pixels. The processing unit counts the number of clustered connected change pixels. When the number of change pixels exceeds the recognition threshold, the processing unit marks the clustered connected change pixels as a picture change area. The frame difference method has a simple algorithm and a fast calculation speed. The calculation pressure of the processing unit is small, and it can have a fast response speed. It can detect dynamic changes in the picture in real time and has a good detection effect on picture changes.

[0028] Furthermore, the aging factor Obtained by the following steps:

[0029] After the equipment completes the maintenance task corresponding to the repair work order, the risk assessment model re-records the equipment's operating time ti. The risk assessment model obtains the design life Tmax of the equipment model through the network. At the same time, it obtains the overall failure rate β of the remaining equipment with the same model in the entire power grid, as well as the maintenance impact caused by the kth failure of the same model equipment. , the value of k is greater than or equal to 1, the maintenance impact The value range is 1-10. The larger the value, the greater the maintenance impact. Determine the specific value based on the duration of on-site equipment maintenance and the number of participants;

[0030] The processing unit of the transfer base station is responsible for statistics of the on-site equipment maintenance duration and the number of participants. Since the operation terminal corresponds to the operator one by one, the processing unit determines the number of participants by recording the number of operation terminals participating in the equipment maintenance. The processing unit records the termination time of the feedback signal sent back by the feedback module of each operation terminal participating in the equipment maintenance. The processing unit subtracts the start time of the operation of the corresponding operation terminal from the termination time of the feedback signal to obtain the duration of the equipment maintenance. The processing unit averages the duration of all operation terminals to obtain the on-site equipment maintenance duration and transmits it to the risk assessment model through the transmission unit.

[0031] The risk assessment model is based on the formula Calculate the aging factor of the equipment , where n is the total number of failures of the equipment, k≤n.

[0032] Furthermore, the specific steps of constructing the inspection process model of the first shadow model are as follows:

[0033] The first shadow model establishes an environment collection library. The first shadow model classifies the received operating environment. The operating environment classification is specifically different types of electrical equipment in the same level of the power grid. Specified environmental labels are generated for different types of operating links. The environmental labels and corresponding operating environments are saved in the environment collection library for subsequent construction of the inspection process model.

[0034] The first shadow model edits the inspection track according to the inspection order of each operation terminal, and the environment tags in the environment collection library are edited according to the operation order. Each operation terminal corresponds to an inspection track. The first shadow model inputs all inspection tracks into the map range of the area where the transfer base station is located;

[0035] The first shadow model uses the PyTorch-DRL4VRP path algorithm to build an inspection process model. The PyTorch-DRL4VRP path algorithm is based on the PyTorch deep reinforcement learning framework and uses the Deep Q-Networks deep reinforcement learning algorithm to calculate the optimal solution for multiple inspection trajectories. The first shadow model re-plans the inspection process model each time it receives a new working environment.

[0036] Furthermore, the specific steps of constructing the maintenance tree model of the second shadow model are as follows:

[0037] The second shadow model establishes an action set library and classifies the received operation actions. The second shadow model realizes classification by calculating the overlap rate between two operation actions. When the overlap rate of two operation actions reaches 95%, the second shadow model marks the two operation actions as the same operation action. The operation action classification is specifically the steps for repairing the same model of equipment. For different types of operation actions, the specified action codes are generated and compiled into a maintenance chain according to the order of the equipment maintenance steps.

[0038] The second shadow model merges the same action codes in all maintenance chains. The second shadow model marks the merged action codes as branch nodes. All maintenance chains are interconnected through branch nodes to obtain a maintenance tree model. Each time the second shadow model receives a new work action, it converts the newly received work action into an action code and merges it into the maintenance tree model, and then updates the maintenance tree model.

[0039] A power grid information operation violation risk detection and control device comprises an operation terminal, a cloud service cluster and a transfer base station. The operation terminal is composed of a recorder, a feedback module, a communication module and a loudspeaker. The operation terminal is installed on the safety helmet of the operator's head or the chest of the operation uniform. The recorder obtains video information of the area where the operator is located. The feedback module obtains a feedback signal of the end of the operation. The loudspeaker is used to play the command voice for the on-site operator to listen to. The recorder is a video recorder of at least two cameras. The feedback module is a physical button or a microphone that recognizes voice commands. The on-site operator triggers the feedback module to obtain a feedback signal by manually pressing the physical button or orally speaking the voice command "completed" or "found a problem" to the microphone. Specifically, the voiceprint data of "completed" or "found a problem" is preset inside the operation terminal. After the sound information collected by the microphone is compressed into a common audio format such as mp3 by the operation terminal, the operation terminal compares the built-in voiceprint data with the compressed sound information. When the similarity is greater than or equal to the comparison threshold, the operation terminal triggers the feedback signal by default. The operation terminal is internally integrated with a battery for power supply.

[0040] Furthermore, the cloud service cluster includes a risk assessment model, a prediction model, a first shadow model, a second shadow model and a repair background. The risk assessment model is used to perform risk assessment on each device and formulate an inspection sequence. The prediction model is used to predict the operator's work steps in advance and send a job prompt for the next step. The first shadow model and the second shadow model optimize subsequent work processes based on the characteristic data generated during the operation of each operator. The repair background is used to collect all repair work orders and issue server instructions for designated equipment repairs. The cloud service cluster is a cloud computing center composed of several cloud servers, and an operating environment is built internally for model training and calculation.

[0041] Furthermore, the transfer base station includes a processing unit and a transmission unit. The transfer base station establishes data communication with the cloud service cluster and the operation terminal respectively through the transmission unit. The processing unit is used for data calculation and processing. The data processed by the processing unit includes video information, server instructions and feedback signals.

[0042] The present invention has the following beneficial effects:

[0043] 1. Using multiple transfer base stations to achieve short-distance communication with the operation terminal can, on the one hand, reduce the computing and bandwidth pressure of the server cluster, and on the other hand, improve the response speed, improve the real-time nature of risk detection, and effectively remind and guide on-site operators to perform operations.

[0044] 2. By constructing an inspection process model, the inspection sequence of power grid equipment can be automatically planned to avoid potential safety risks of the equipment. By constructing a maintenance tree model, the operator's work steps can be predicted in advance and reminders can be given to avoid safety accidents caused by operational errors or out-of-scope operations.

[0045] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0047] Figure 1 This is a system block diagram of a power grid information operation violation risk detection and control device of the present invention;

[0048] Figure 2 The present invention is a flow chart of a method for detecting and controlling risk of power grid information operations violations. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1-2The present invention provides a technical solution: a method and device for detecting and controlling the risk of illegal operation of power grid information, which is implemented based on a cloud service cluster, a transfer base station and an operation terminal. The cloud service cluster includes a risk assessment model, a prediction model, a first shadow model, a second shadow model and a repair background. The transfer base station includes a processing unit and a transmission unit. The operation terminal includes a recorder, a feedback module, a communication module and a speaker. The control method includes an inspection program and a repair program, such as Figure 2 As shown, the inspection procedure includes the following steps:

[0051] The operator inspects the power grid equipment within the signal coverage of the transfer base station. The recorder obtains the video information of the inspection in real time. The video information is transmitted to the processing unit through the communication module and the transmission unit in turn. The processing unit and the first shadow model cooperate to execute the video feature extraction program to obtain feature data. The feature data includes the operator's operation action and the operator's working environment. The first shadow model constructs the inspection process model based on the feature data. The inspection process model continuously obtains new feature data for automatic iteration and update.

[0052] When the operator finds a fault or safety defect in the power grid equipment during the inspection task, the feedback module is triggered actively. The feedback signal of the feedback module is transmitted to the nearest transfer base station processing unit through the communication module and the transmission unit in turn. The processing unit generates a repair work order and transmits it to the repair backend in the cloud service cluster through the transmission unit. After the operator completes the repair procedure, the next step is executed;

[0053] The risk assessment model presets the aging coefficient of the equipment corresponding to each repair work order , based on the aging factor Conduct risk assessment on each device, specifically adjust the inspection weight of each device , the calculation formula is , γ is the number of devices of this model within the signal coverage of the transfer base station, and the risk assessment model is based on the inspection weight of each device in the inspection process model Update the inspection order of the equipment. The inspection order is based on the inspection weight. Arranged from large to small, inspection power is great Prioritize the inspection of equipment. By continuously optimizing the inspection sequence of equipment, equipment that may pose safety risks in the power grid can be detected in advance, making it easier for operators to report maintenance orders in a timely manner.

[0054] The repair process includes the following steps:

[0055] The repair backend collects repair work orders and sends server instructions to the processing unit of the corresponding transfer base station. The processing unit sends the maintenance task corresponding to the server instruction to the corresponding operation terminal through the transmission unit. The operator corresponds to the operation terminal one by one, and the operator is assigned to the location corresponding to the maintenance task to perform the operation;

[0056] When the operator performs maintenance work, the processing unit and the second shadow model work together to execute the video feature extraction program to obtain feature data, and the second shadow model builds a maintenance tree model based on the feature data. The maintenance tree model continuously obtains new feature data for automatic iterative update;

[0057] The prediction model predicts the operator's work steps in advance according to the maintenance tree model. Before the operator's current work is completed, the prediction model sends the next step of the work prompt. Specifically, the processing unit transmits the operator's characteristic data to the prediction model through the transmission unit. The prediction model sends the server instruction to the processing unit according to the maintenance tree model. The processing unit transmits the server instruction to the speaker through the transmission unit and the communication module in turn. The speaker converts the server instruction into audio playback for the operator to listen to. The prediction model has built-in power safety operation procedures of the power grid company. Combined with the maintenance tree model, it predicts the risks that may be caused by the operator's work one step in advance, and sends the safety steps that need to be paid attention to to the corresponding transfer base station processing unit in the form of server instructions. The processing unit converts the server instruction into a prompt voice and forwards it to the operation terminal through the transmission unit. The speaker of the operation terminal broadcasts it, which can indirectly improve the safety awareness of the on-site operators. At the same time, the above steps can also identify the on-site operators who have performed out-of-scope operations, that is, the operation steps that exceed the power grid company's power safety operation procedures. The speaker of the operation terminal also broadcasts a warning voice to remind the on-site operators to pay attention.

[0058] It should be noted that if the operation terminal directly communicates data with the server cluster, and the operation terminal is mainly for the video information obtained by the recorder, on the one hand, it will increase the computing pressure and communication bandwidth pressure of the server cluster, and there may be delays in the issuance of server instructions. On the other hand, excessive data transmission is likely to lead to unstable data transmission, while increasing the battery life and heat generation of the operation terminal. Therefore, using multiple relay base stations for signal coverage can achieve short-distance real-time data communication. The processing unit filters the video information to extract feature data to alleviate the pressure of data communication. At the same time, using multiple processing units for distributed computing can reduce the computing pressure of the server cluster, allowing the server cluster to focus on inspection program optimization and maintenance step prediction. Short-distance communication between the relay base station and the operation terminal can improve response speed, improve the real-time nature of risk detection, and avoid operators waiting on site. The prediction model sends server instructions at least 2-3 steps in advance. After receiving the processor instruction, the processing unit can send the next step to the operation terminal according to the received feedback signal, avoiding the delay problem in the issuance of server instructions.

[0059] Among them, the video feature extraction procedure includes the following steps:

[0060] The processing unit fuses multiple video information to obtain a panoramic video, and the processing unit extracts operation features from the panoramic video;

[0061] Human action recognition: The processing unit uses the OpenPose algorithm detection library to extract the joint nodes of the person from the panoramic video. If the algorithm detection library does not extract the joint nodes of the operator from the panoramic video, it jumps to the working environment recognition step. Otherwise, it proceeds to the next step.

[0062] The joint nodes include the head, torso, left upper arm, left lower arm, left hand, right upper arm, right lower arm, right hand, left thigh, left calf, left foot, right thigh, right calf and right foot. The processing unit uses line segments to connect the joint nodes of the character to obtain a simplified skeleton. The processing unit extracts the simplified skeleton of each frame in the panoramic video and arranges and plays them in the order of the previous and next frames to obtain a simplified action. The processing unit transmits the simplified action to the first shadow model in the cloud service cluster through the transmission unit. The first shadow model uses the YOLOv8 image recognition algorithm to identify the operation action corresponding to the simplified action. The processing unit uses the simplified skeleton as the center line and extends a preset width of 50 pixels to both sides of the center line to obtain a widened rectangle. The processing unit removes the widened rectangle from the panoramic video and proceeds to the next step.

[0063] Working environment recognition: the processing unit uses a frame difference method to identify the picture change area in the panoramic video. The processing unit transmits the picture change area to the first shadow model through the transmission unit. The first shadow model integrates the training data set. The first shadow model is trained based on the YOLOv8 image recognition algorithm using the training data set. The first shadow model uses the image recognition algorithm to identify the working environment corresponding to the picture change area.

[0064] It should be noted that the training data sets are the irregular behavior detection data set of substation operators, the transmission line insulator defect data set, the insulating oil leakage detection image data set inside the power equipment, the GIS partial discharge defect detection data set, the distribution room status detection data set, the transmission line foreign body data set and the distribution power component defect data set. The above training data sets all use VOC labels and can be directly used for the training of the YOLOv8 image recognition algorithm.

[0065] The method for obtaining the panoramic video specifically includes the following steps:

[0066] Time axis synchronization: when the recorder starts recording, the recording time will be displayed in the upper right corner of the video information screen. The processing unit will synchronize the time axes of multiple video information based on the recording time and enter the next processing step;

[0067] Screen alignment: the processing unit converts the first frame of the video information into a first grayscale image, the grayscale value range of the first grayscale image is 0-255, the processing unit calculates the grayscale value difference between the pixels in the first grayscale image, normalizes the grayscale value difference of all pixels, assigns the maximum grayscale value difference to 1, assigns the minimum grayscale value difference to 0, and converts the remaining grayscale value differences in geometric proportion to the maximum and minimum values, marks the pixels with grayscale value differences greater than the judgment threshold of 0.5 as contour lines, and aligns the contour lines in all video information to obtain a fused video, and proceeds to the next processing step;

[0068] De-noising, the processing unit uses Gaussian filtering to de-noise the fused video and proceeds to the next processing step;

[0069] Distortion correction,The processing unit uses the FishEye camera model to perform distortion correction on the fused video to obtain a panoramic video.

[0070] Among them, the specific steps of the frame difference method to identify the image change area in the panoramic video are as follows:

[0071] The processing unit converts each frame of the panoramic video into a second grayscale image. The grayscale value range of the second grayscale image is 0-127. The identification of the image change area only requires simple identification, and there is no need to refine the grayscale value range. The grayscale value range of 0-127 can meet the identification requirements, which can reduce a certain amount of calculation and speed up the response speed of the processing unit. The processing unit calculates the grayscale value difference between pixels in the second grayscale image, and normalizes the grayscale value difference values ​​of all pixels. The normalization process is consistent with the normalization process of the first grayscale image.

[0072] The processing unit sets the change threshold to 0.28, and marks the pixels in the second grayscale image whose adjacent pixels are greater than the change threshold as changed pixels. The processing unit counts the number of clustered connected changed pixels. When the number of changed pixels exceeds the recognition threshold of 50, the processing unit marks the clustered connected changed pixels as a picture change area. The frame difference method has a simple algorithm and a fast calculation speed. The calculation pressure of the processing unit is small, and it can have a fast response speed. It can detect dynamic changes in the picture in real time and has a good detection effect on picture changes.

[0073] Among them, the aging coefficient Obtained by the following steps:

[0074] After the equipment completes the maintenance task corresponding to the repair work order, the risk assessment model re-records the equipment's operating time ti. The risk assessment model obtains the design life Tmax of the equipment model through the network. At the same time, it obtains the overall failure rate β of the remaining equipment with the same model in the entire power grid, as well as the maintenance impact caused by the kth failure of the same model equipment. , the value of k is greater than or equal to 1, the maintenance impact The value range is 1-10. The larger the value, the greater the maintenance impact. Determine the specific value based on the duration of on-site equipment maintenance and the number of participants;

[0075] The processing unit of the transfer base station is responsible for statistics of the on-site equipment maintenance duration and the number of participants. Since the operation terminal corresponds to the operator one by one, the processing unit determines the number of participants by recording the number of operation terminals participating in the equipment maintenance. The processing unit records the termination time of the feedback signal sent back by the feedback module of each operation terminal participating in the equipment maintenance. The processing unit subtracts the start time of the operation of the corresponding operation terminal from the termination time of the feedback signal to obtain the duration of the equipment maintenance. The processing unit averages the duration of all operation terminals to obtain the on-site equipment maintenance duration and transmits it to the risk assessment model through the transmission unit.

[0076] The risk assessment model is based on the formula Calculate the aging factor of the equipment , where n is the total number of failures of the equipment, k≤n.

[0077] Among them, the specific steps of the first shadow model to build the inspection process model are as follows:

[0078] The first shadow model establishes an environment collection library. The first shadow model classifies the received operating environment. The operating environment classification is specifically different types of electrical equipment in the same level of the power grid. Specified environmental labels are generated for different types of operating links. The environmental labels and corresponding operating environments are saved in the environment collection library for subsequent construction of the inspection process model.

[0079] The first shadow model edits the inspection track according to the inspection order of each operation terminal, and the environment tags in the environment collection library are edited according to the operation order. Each operation terminal corresponds to an inspection track. The first shadow model inputs all inspection tracks into the map range of the area where the transfer base station is located;

[0080] The first shadow model uses the PyTorch-DRL4VRP path algorithm to build an inspection process model. The PyTorch-DRL4VRP path algorithm is based on the PyTorch deep reinforcement learning framework and uses the Deep Q-Networks deep reinforcement learning algorithm to calculate the optimal solution for multiple inspection trajectories, thereby achieving the most efficient inspection within the map range of the transit base station area. Each time the first shadow model receives a new operating environment, it re-plans the inspection trajectory in the inspection process model to achieve automatic updating of the inspection process model.

[0081] Among them, the specific steps of constructing the maintenance tree model of the second shadow model are as follows:

[0082] The second shadow model establishes an action set library and classifies the received operation actions. The second shadow model realizes classification by calculating the overlap rate between two operation actions. When the overlap rate of two operation actions reaches 95%, the second shadow model marks the two operation actions as the same operation action. The operation action classification is specifically the steps for repairing the same model of equipment. For different types of operation actions, the specified action codes are generated and compiled into a maintenance chain according to the order of the equipment maintenance steps.

[0083] The second shadow model merges the same action codes in all maintenance chains. The second shadow model marks the merged action codes as branch nodes. All maintenance chains are interconnected through branch nodes to obtain a maintenance tree model. Each time the second shadow model receives a new work action, it converts the newly received work action into an action code and merges it into the maintenance tree model, and then updates the maintenance tree model.

[0084] A power grid information operation violation risk detection and control device, such as Figure 1As shown, it includes an operation terminal, a cloud service cluster and a transfer base station. The operation terminal is composed of a recorder, a feedback module, a communication module and a loudspeaker. The operation terminal is installed on the operator's safety helmet or the chest of the operation uniform. The recorder obtains video information of the operator's area, the feedback module obtains a feedback signal of the end of the operation, and the loudspeaker is used to play the command voice for the on-site operator to listen to. The recorder is a video recorder of at least two cameras, and the feedback module is a physical button or a microphone that recognizes voice commands. The on-site operator triggers the feedback module to obtain a feedback signal by manually pressing a physical button or orally speaking a voice command "completed" or "found a problem" to the microphone. Specifically, the voiceprint data of "completed" or "found a problem" is preset in the operation terminal. After the sound information collected by the microphone is compressed into common audio formats such as mp3 by the operation terminal, the operation terminal compares the built-in voiceprint data with the compressed sound information. When the similarity is greater than or equal to the comparison threshold of 90%, the operation terminal triggers the feedback signal by default. A battery is integrated in the operation terminal for power supply.

[0085] Among them, the cloud service cluster includes a risk assessment model, a prediction model, a first shadow model, a second shadow model and a repair background. The risk assessment model is used to perform risk assessment on each device and formulate an inspection sequence. The prediction model is used to predict the operator's work steps in advance and send the next step of the work prompt. The first shadow model and the second shadow model optimize the subsequent work process based on the characteristic data generated during the operation of each operator. The repair background is used to collect all repair work orders and issue server instructions for the maintenance of designated equipment. The cloud service cluster is a cloud computing center composed of several cloud servers. An operating environment for model training and calculation is built internally. The specific model of the cloud server is not limited here, and it only needs to meet the computing power requirements and communication bandwidth of the server cluster.

[0086] Among them, the transfer base station includes a processing unit and a transmission unit. The transfer base station establishes data communication with the cloud service cluster and the operation terminal respectively through the transmission unit. The processing unit is used for data calculation and processing. The data processed by the processing unit includes video information, server instructions and feedback signals.

[0087] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting and controlling the risk of illegal operation of power grid information, including an inspection procedure and a repair procedure, characterized in that: The inspection procedure includes the following steps: The recorder obtains video information during inspection in real time, executes the video feature extraction program to obtain feature data, and the first shadow model constructs the inspection process model, continuously obtains feature data for automatic update; When the operator finds a fault or safety defect during the inspection task, the feedback module is triggered to generate a repair work order and transmit it to the repair backend. After the repair procedure is completed, the next step is executed; The risk assessment model presets an aging coefficient for each device, performs risk assessment on each device, and updates the inspection sequence; The repair process includes the following steps: The repair backend collects repair work orders and sends server instructions to the operation terminal; When the operator performs maintenance work, the video feature extraction program is executed to obtain feature data, and the second shadow model constructs a maintenance tree model, continuously obtains feature data for automatic update; The prediction model predicts the operator's work steps in advance, sends the next step of the work prompt, and identifies when the on-site operator performs work beyond the scope, and the work terminal speaker broadcasts a warning voice.

2. A method for detecting and controlling risk of power grid information operation violations according to claim 1, characterized in that: The video feature extraction procedure includes the following steps: Fusion of multiple video information to obtain panoramic video, and extraction of operation features from the panoramic video; Human action recognition, using the algorithm detection library to extract the joint nodes of the person from the panoramic video. If the algorithm detection library does not extract the joint nodes of the operator from the panoramic video, it jumps to the working environment recognition step, otherwise, it proceeds to the next step; Use line segments to connect the joint nodes of the character to obtain a simplified skeleton, extract the simplified skeleton of each frame in the panoramic video and arrange and play them in the order of the previous and next frames to obtain a simplified action, transmit the simplified action to the first shadow model, use the image recognition algorithm to identify the work action corresponding to the simplified action, use the simplified skeleton as the center line, extend the preset width to both sides of the center line to obtain a widened rectangle, remove the widened rectangle from the panoramic video, and proceed to the next step; Working environment recognition, the processing unit uses a frame difference method to identify the area where the picture changes in the panoramic video, and transmits the area where the picture changes to the first shadow model. The first shadow model integrates a training data set. The first shadow model is trained based on an image recognition algorithm using the training data set, and uses the image recognition algorithm to identify the working environment corresponding to the area where the picture changes.

3. A method for detecting and controlling risk of power grid information operation violations according to claim 2, characterized in that: The method for obtaining the panoramic video specifically includes the following steps: When the recorder starts recording, the recording time will be displayed in the upper right corner of the video information screen. The time axis of multiple video information will be synchronized based on the recording time and enter the next processing step; The first frame of the video information is converted into a first grayscale image, with a grayscale value range of 0-255, the grayscale value difference between pixels in the first grayscale image is calculated, the grayscale value difference of all pixels is normalized, the pixels with a grayscale value difference greater than a determination threshold are marked as contour lines, the contour lines in all video information are aligned to obtain a fused video, and the next processing step is performed; Use Gaussian filtering to remove noise from the fused video and proceed to the next processing step; The camera model is used to perform distortion correction on the fused video to obtain a panoramic video.

4. A method for detecting and controlling risk of power grid information operation violations according to claim 2, characterized in that: The specific steps of the frame difference method to identify the image change area in the panoramic video are as follows: The processing unit converts each frame of the panoramic video into a second grayscale image with a grayscale value range of 0-127, calculates the grayscale value difference between pixels in the second grayscale image, and normalizes the grayscale value differences of all pixels; A change threshold is set, and pixels whose adjacent pixels in the second grayscale image are greater than the change threshold are marked as change pixels. The number of clustered connected change pixels is counted. When the number of change pixels exceeds the recognition threshold, the clustered connected change pixels are marked as a picture change area.

5. A method for detecting and controlling risk of power grid information operation violations according to claim 1, characterized in that: Aging factor Obtained by the following steps: After the equipment completes the repair work order, the risk assessment model re-records the equipment's operating time ti. The risk assessment model obtains the design life Tmax of the equipment model, and also obtains the overall failure rate β of the remaining equipment with the same model as the equipment, as well as the maintenance impact caused by the kth failure of the same model equipment. , maintenance impact The value range is 1-10. The larger the value, the greater the maintenance impact. Determine the specific value based on the duration of on-site equipment maintenance and the number of participants; The processing unit is responsible for counting the duration of on-site equipment maintenance and the number of participants. The number of participants is determined by the number of operating terminals, and the termination time of the feedback signal of each operating terminal is recorded. The duration of equipment maintenance is obtained by subtracting the start time of the operation from the termination time of the feedback signal. The duration of all operating terminals is averaged to obtain the on-site equipment maintenance duration and transmitted to the risk assessment model. The risk assessment model is based on the formula Calculate the aging factor of the equipment , where n is the total number of failures of the device.

6. A method for detecting and controlling risk of power grid information operation violations according to claim 1, characterized in that: The specific steps for constructing the inspection process model of the first shadow model are as follows: Establish an environment collection library, classify the received working environments, generate designated environment tags for different types of working links, and save the environment tags and corresponding working environments to the environment collection library; According to the inspection order of each operating terminal, the environment tags in the environment collection library are edited into inspection tracks according to the operation order. Each operating terminal corresponds to an inspection track, and all inspection tracks are input into the map range of the area where the transfer base station is located; The inspection process model is constructed using the path algorithm. Each time the first shadow model receives a new operating environment, it replans the inspection process model.

7. A method for detecting and controlling risk of power grid information operation violations according to claim 1, characterized in that: The specific steps of constructing the maintenance tree model of the second shadow model are as follows: Establish an action set library to classify the received operation actions. The operation action classification is specifically the steps for repairing the same type of equipment. Generate specified action codes for different types of operation actions and edit them into a maintenance chain in the order of equipment maintenance steps. The same action codes in the maintenance chain are merged, and the merged action codes are marked as branch nodes. All maintenance chains are connected to each other through branch nodes to obtain a maintenance tree model. Each time the second shadow model receives a new operation action, it converts the operation action into an action code and merges it into the maintenance tree model, and then updates the maintenance tree model.

8. A power grid information operation violation risk detection and control device, characterized in that A method for detecting and controlling the risk of illegal operations of power grid information operations is applied to implement any one of claims 1-7, comprising an operation terminal, a cloud service cluster and a transfer base station, wherein the operation terminal is composed of a recorder, a feedback module, a communication module and a speaker, and the operation terminal is installed on the operator's head helmet or the chest of the operation uniform, the recorder obtains video information of the area where the operator is located, the feedback module obtains feedback signals, the speaker is used to play command voice, the recorder is a video recorder of at least two cameras, the feedback module is a physical button or a microphone that recognizes voice commands, and the feedback module is triggered by manually pressing the physical button or speaking the voice command "completed" or "found problem" to the microphone, and a battery is integrated in the operation terminal for power supply.

9. A power grid information operation violation risk detection and control device according to claim 8, characterized in that: The cloud service cluster includes a risk assessment model, a prediction model, a first shadow model, a second shadow model and a repair background. The risk assessment model is used to perform risk assessment on each device and formulate an inspection sequence. The prediction model is used to predict the operator's work steps in advance and send a job prompt for the next step. The first shadow model and the second shadow model optimize the subsequent work process based on the characteristic data generated during the operation of each operator. The repair background is used to collect all repair work orders and issue specified server instructions.

10. A power grid information operation violation risk detection and control device according to claim 8, characterized in that: The transfer base station includes a processing unit and a transmission unit. The transfer base station establishes data communication with the cloud service cluster and the operation terminal respectively through the transmission unit. The processing unit is used for data calculation and processing.

Images