A Smart Error Prevention Method Based on Cloud-Edge Collaboration and Image Recognition

Abstract

This invention relates to an intelligent error prevention method based on cloud-edge collaboration and image recognition, comprising: establishing a rule base: the rule base includes an electrical rule base and a management rule base, the establishment of which is the foundation for error prevention, and the rule base is used for error prevention judgment by the cloud decision system and the edge decision system; performing power grid topology analysis; constructing a cloud-edge decision system; and performing image recognition for the edge decision system, where image recognition provides auxiliary criteria for decision-making. This invention starts from the source, obtains the complete source flow path chain and its spanning tree, and obtains error prevention edge zones by dividing the tree. Multiple edge nodes are then set for each edge zone to maintain the monitored equipment area. This solves the drawbacks of the lack of rationality in block division and the concentration of data processing on a single device in error prevention methods; relying on enhanced image recognition technology as an auxiliary technology for error prevention judgment increases the accuracy of error prevention judgment.

Description

Technical Field

[0001] This invention relates to the field of substation anti-misoperation technology, and in particular to an intelligent anti-misoperation method based on cloud-edge collaboration and image recognition. Background Technology

[0002] Misoperation prevention technology is a crucial safety guarantee for modern intelligent substations. Misoperation can lead to accidents in substations, and these accidents can cause widespread line faults through transmission lines. Therefore, the necessity of researching misoperation prevention methods is self-evident.

[0003] Preventing accidental operation requires accurate network topology and device status information. However, using undirected network graphs for single topology analysis can lead to multiple data searches and untimely acquisition of device information. Preventing accidental operation also necessitates the establishment of a decision-making system; however, traditional decision-making often relies on a single edge decision-making server, which can lead to server overload and delayed operation processing. Furthermore, traditional error prevention methods neglect the importance of auxiliary criteria, failing to incorporate video surveillance into the error prevention judgment process and lacking necessary supplementary elements. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent error prevention method based on cloud-edge collaboration and image recognition that rationally divides blocks, effectively alleviates the pressure on data access in cloud decision-making systems, and enhances the accuracy of error prevention judgment.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent error prevention method based on cloud-edge collaboration and image recognition, the method comprising the following sequential steps:

[0006] (1) Establishing a rule base: The rule base includes an electrical rule base and a management rule base. The establishment of the rule base is the basis for preventing misoperation. The rule base is used for the judgment of preventing misoperation in cloud decision-making system and edge decision-making system;

[0007] (2) Perform power grid topology analysis;

[0008] (3) Construct a cloud-edge decision-making system;

[0009] (4) Perform image recognition for the edge decision system. Image recognition provides auxiliary criteria for the edge decision system to make decisions.

[0010] The step (1) specifically refers to: by applying electrical rules and management rules, sorting out the operation rules for equipment and their operation modes of circuit breakers, disconnect switches, grounding switches, grid / cabinet doors; the establishment of the electrical rule library is based on the electrical specifications of substations, and the management rule library is established according to the actual and frequent fault situations of each substation.

[0011] Step (2) specifically includes the following steps:

[0012] (2a) Constructing a directed graph: Abstract the nodes and branches of the power network as vertices and edges respectively, and take the direction of active power flow in this state to obtain the directed graph of the power network. In this context, a power grid node is abstracted as a set of vertices. The set of power transmission branches is abstracted as element arc , For element arc The starting point For element arc The endpoint, from which all sources are obtained, including incoming lines and buses;

[0013] (2b) Choose any source as the starting point and mark it as searched;

[0014] (2c) Establish the adjacency list of the sending node and the power transmission branch, and the vertex set of the directed graph. The set of sending nodes and the set of vertices in the adjacency list of the sending node-power transmission branch. nodes Corresponding to nodes All power transmission branches ;

[0015] (2d) Perform the first round of topology analysis: using the established sender node-power transmission branch adjacency table, perform topology processing based on active power flow;

[0016] (2e) Perform a second round of topology analysis to obtain the complete source-flow path chain;

[0017] (2f) The purpose of traversal search is achieved by using a layer-by-layer approach;

[0018] (2g) Repeat steps (2c) to (2f) until all nodes have been traversed.

[0019] The step (3) specifically refers to: the cloud-edge decision system includes a cloud decision system and an edge decision system. Both the cloud decision system and the edge decision system have cloud-edge interaction channels, rule bases, data acquisition and integration modules, data processing modules, instruction issuance modules and network topology modules. The edge decision system also has an image recognition module, and the cloud decision system also has a module for setting the priority of the edge decision system.

[0020] The cloud decision-making system and the edge decision-making system are deployed on cloud servers and on-site edge servers, respectively. The anti-misoperation host is used as the decision-making implementer, and the anti-misoperation edge zone and anti-misoperation edge nodes are used as the data sources for the cloud decision-making system and the edge decision-making system. The anti-misoperation host processes the data and images, and the anti-misoperation edge zone divides the obtained topology spanning tree based on the line where the substation equipment is located. It stores and applies the equipment data and video stream information sent by the anti-misoperation edge nodes. Video monitoring equipment and data acquisition equipment are deployed at the anti-misoperation edge nodes to collect video stream images and equipment status information and transmit them wirelessly to neighboring nodes and the anti-misoperation edge zone.

[0021] Set one This flag indicates that when communication between the cloud decision system and the edge decision system is normal, this value is 1; otherwise, it is set to 0. When communication is normal, the edge decision system collects data from the error-prevention edge zone and labels each edge zone as follows: ,Will The corresponding data is processed and analyzed, and then sent to the cloud decision-making system; when communication is normal, that is... At that time, the error prevention host receives error prevention instructions from the cloud decision system and sends them to the corresponding devices; when At that time, the side decision-making system directly sends the error prevention information to the error prevention host;

[0022] When the cloud decision system needs to process more data than the threshold or the concurrent requests of the edge decision system reach their limit, the cloud decision system will frequently send data to the edge region or areas with large data volumes. The data is retrieved and fed back to the corresponding edge. The edge decision-making system first compares it with the cloud decision-making system. The transmission channel is closed and handled automatically by the edge; if the result is a rejection operation, it is sent to the anti-error host after processing for simulation; if the result is an executable operation, the cloud will open the corresponding channel when the cloud decision system is idle. At this point, the edge decision system provides data to the cloud again, and the cloud decision system confirms whether there is a situation involving edge lock mutual exclusion.

[0023] Step (4) specifically includes the following steps:

[0024] (4a) Gaussian mixture modeling:

[0025] (4a1) Install video surveillance equipment in fixed locations and set up video patrol machines in important areas to obtain clear image data;

[0026] (4a2) The edge performs background modeling based on the obtained video images, and the pixel time series is represented as follows:

[0027]

[0028] In the formula, Indicates time grayscale value, Pixels at any moment The probability calculation formula is:

[0029]

[0030] In the formula, for Time of the first Each model weight value; for Time of the first The mean of a Gaussian distribution; for Covariance at time; for The probability density function at time t is calculated using the following formula:

[0031]

[0032] (4a3) Parameter update:

[0033]

[0034] In the formula, For learning rate, For threshold coefficient, For Time of the first The variance of a Gaussian distribution;

[0035] The grayscale values ​​are sorted according to priority by Gaussian distribution and checked whether they satisfy equation (1). If they satisfy the Gaussian distribution, it means that the grayscale value matches the Gaussian distribution, and the first value that satisfies the Gaussian distribution is updated in the order of equation (2) to (5). If they do not satisfy the Gaussian distribution, the data is updated in the order of equation (6).

[0036] (4a4) Target extraction:

[0037] After the Gaussian model of the video image is established, according to Sort in descending order, starting with the first few rows. A Gaussian distribution is used as the background:

[0038]

[0039] In the formula, The threshold is used to match the current pixel with the obtained background. If the match fails, the pixel is considered a moving target.

[0040] (4b) Image recognition:

[0041] The Faster R-CNN network was used as an image recognition tool. The Faster R-CNN network was trained as follows:

[0042] (4b1) Organize the extracted targets into a test set, and create a dataset after target extraction, and then label the dataset accordingly;

[0043] (4b2) Input the dataset into the Faster R-CNN network for parameter fine-tuning and adaptive learning. Use the region generation network in the Faster R-CNN network architecture to extract useful anchor information. On this basis, perform bounding box regression to improve the target grasping accuracy and obtain the updated weights and pooling parameters of each convolutional layer of the Faster R-CNN network.

[0044] (4b3) Save the obtained data of each layer of the Faster R-CNN network, and use the test set to identify the number of people and whether there is any accidental entry within the perimeter through the Faster R-CNN network trained in step (4b2): When there is a moving target in the restricted area, it indicates that the energized interval has been entered by accident. The edge decision system directly issues an anti-mistake instruction to the anti-mistake host, and the edge decision system transmits the image data to the cloud decision system for remote control; when the detected target is only one person, it indicates that the operation is performed by a single person. The edge decision system directly issues an anti-mistake instruction to the anti-mistake host, and the edge transmits the image data to the cloud decision system for remote control.

[0045] As can be seen from the above technical solution, the beneficial effects of the present invention are as follows: First, the present invention starts from the source, obtains a complete source flow path chain and its spanning tree, and obtains the anti-misoperation edge zone by dividing the tree. Then, multiple edge nodes are set for each edge zone to maintain the monitored device area. This solves the drawbacks of the lack of rationality in block division and the concentration of data processing on a single device in the anti-misoperation method. Second, the present invention uses a cloud-edge system processing method, which not only distributes the device's data information to each edge decision system, but also gives the cloud decision system sufficient authority to process the data from the edge decision system. At the same time, considering the excessive concurrent tasks and data overload, the edge decision system is given a certain degree of autonomous anti-misoperation authority, which effectively alleviates the data access pressure of the cloud decision system. Third, relying on enhanced image recognition technology as an auxiliary technology for anti-misoperation judgment increases the accuracy of anti-misoperation judgment. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the intelligent error prevention logic architecture of the present invention;

[0047] Figure 2 This is a flowchart of the method of the present invention;

[0048] Figure 3 Flowchart for error prevention and auxiliary verification of image recognition. Detailed Implementation

[0049] like Figure 2 As shown, an intelligent error prevention method based on cloud-edge collaboration and image recognition is proposed. This method includes the following steps in sequence:

[0050] (1) Establishing a rule base: The rule base includes an electrical rule base and a management rule base. The establishment of the rule base is the basis for preventing misoperation. The rule base is used in the judgment of preventing misoperation in cloud decision-making system and edge decision-making system;

[0051] (2) Perform power grid topology analysis: transform complex wiring diagrams into simple graph relationships, obtain topology spanning trees using graph relationships, and use the branches and number of branch nodes of the topology spanning tree as one of the bases for setting up anti-misoperation edge nodes and the number of video inspection equipment and monitoring equipment.

[0052] (3) Construct a cloud-edge decision system; The cloud-edge decision system obtains the equipment environment after equipment operation through power grid topology analysis, including the corresponding bay and equipment connection relationship, generates a new logical expression for equipment operation, and compares it with the rule base to obtain the return value;

[0053] (4) Perform image recognition for the side decision system: Image recognition for the side decision system is a supplement to the rule base for error prevention. The safety of substation operation lies not only in the equipment but also in the safety of the target. Image recognition provides auxiliary criteria for the side decision system to make decisions.

[0054] The step (1) specifically refers to: by applying electrical rules and management rules, sorting out the operation rules for equipment and their operation modes of circuit breakers, disconnect switches, grounding switches, grid / cabinet doors; the establishment of the electrical rule library is based on the electrical specifications of substations, and the management rule library is established according to the actual and frequent fault situations of each substation.

[0055] Step (2) specifically includes the following steps:

[0056] (2a) Constructing a directed graph: Abstract the nodes and branches of the power network as vertices and edges respectively, and take the direction of active power flow in this state to obtain the directed graph of the power network. In this context, a power grid node is abstracted as a set of vertices. The set of power transmission branches is abstracted as element arc , For element arc The starting point For element arc The endpoint, from which all sources are obtained, including incoming lines and buses;

[0057] (2b) Choose any source as the starting point and mark it as searched;

[0058] (2c) Establish the adjacency list of the sending node and the power transmission branch, and the vertex set of the directed graph. The set of nodes that are the originating nodes in the adjacency list, and the set of vertices. nodes Corresponding to nodes All power transmission branches ;

[0059] (2d) Perform the first round of topology analysis: using the established sender node-power transmission branch adjacency table, perform topology processing based on active power flow;

[0060] (2e) Perform a second round of topology analysis to obtain the complete source-flow path chain;

[0061] (2f) The purpose of traversal search is achieved by using a layer-by-layer approach;

[0062] (2g) Repeat steps (2c) to (2f) until all nodes have been traversed.

[0063] The step (3) specifically refers to: the cloud-edge decision system includes a cloud decision system and an edge decision system. Both the cloud decision system and the edge decision system have cloud-edge interaction channels, rule bases, data acquisition and integration modules, data processing modules, instruction issuance modules and network topology modules. The edge decision system also has an image recognition module, and the cloud decision system also has a module for setting the priority of the edge decision system.

[0064] The cloud decision-making system and the edge decision-making system are deployed on cloud servers and on-site edge servers, respectively. The anti-misoperation host is used as the decision-making implementer, and the anti-misoperation edge zone and anti-misoperation edge nodes serve as the data sources for the cloud decision-making system and the edge decision-making system. The role of the anti-misoperation host is to process data and images. The anti-misoperation edge zone is to divide the obtained topology spanning tree based on the line where the substation equipment is located. It stores and applies the equipment data and video stream information sent by the anti-misoperation edge nodes. Video monitoring equipment and data acquisition equipment are deployed at the anti-misoperation edge nodes to collect video stream images and equipment status information and transmit them wirelessly to neighboring nodes and the anti-misoperation edge zone at a certain period of time.

[0065] Set one This flag indicates that when communication between the cloud decision system and the edge decision system is normal, this value is 1; otherwise, it is set to 0. When communication is normal, the edge decision system collects data from the error-prevention edge zone and labels each edge zone as follows: ,Will The corresponding data is processed and analyzed accordingly, and then sent to the cloud decision-making system; when communication is normal, that is... At that time, the error prevention host receives error prevention instructions from the cloud decision system and sends them to the corresponding devices; when At that time, the side decision-making system directly sends the error prevention information to the error prevention host;

[0066] When the cloud decision system needs to process more data than the threshold or the concurrent requests of the edge decision system reach their limit, the cloud decision system will frequently send data to the edge region or areas with large data volumes. The data is retrieved and fed back to the corresponding edge. The edge decision-making system first compares it with the cloud decision-making system. The transmission channel is closed and handled automatically by the edge; if the result is a rejection operation, it is sent to the anti-error host after processing for simulation; if the result is an executable operation, the cloud will open the corresponding channel when the cloud decision system is idle. At this point, the edge decision system provides data to the cloud again, and the cloud decision system confirms whether there is a situation involving edge lock mutual exclusion.

[0067] like Figure 3 As shown, step (4) specifically includes the following steps:

[0068] (4a) Gaussian mixture modeling:

[0069] (4a1) Install a certain number of video surveillance devices at fixed locations and set up video patrol machines in important areas to obtain clear image data;

[0070] (4a2) Based on the obtained video images, the edge performs background modeling, and the pixel time series can be represented as:

[0071]

[0072] In the formula, Indicates time grayscale value, Pixels at any moment The probability calculation formula is:

[0073]

[0074] In the formula, for Time of the first Each model weight value; for Time of the first The mean of a Gaussian distribution; for Covariance at time; for The probability density function at time t is calculated using the following formula:

[0075]

[0076] (4a3) Parameter update:

[0077]

[0078] In the formula, For learning rate, For threshold coefficient, For Time of the first The variance of a Gaussian distribution;

[0079] The grayscale values ​​are sorted according to priority by Gaussian distribution and checked whether they satisfy equation (1). If they satisfy the Gaussian distribution, it means that the grayscale value matches the Gaussian distribution, and the first value that satisfies the Gaussian distribution is updated in the order of equation (2) to (5). If they do not satisfy the Gaussian distribution, the data is updated in the order of equation (6).

[0080] (4a4) Target extraction:

[0081] After the Gaussian model of the video image is established, according to Sort in descending order, starting with the first few rows. A Gaussian distribution is used as the background:

[0082]

[0083] In the formula, The threshold is used to match the current pixel with the obtained background. If the match fails, the pixel is considered a moving target.

[0084] (4b) Image recognition:

[0085] The Faster R-CNN network was used as an image recognition tool. The Faster R-CNN network was trained as follows:

[0086] (4b1) Organize the extracted targets into a test set, and create a dataset after target extraction, and then label the dataset accordingly;

[0087] (4b2) Input the dataset into the Faster R-CNN network for parameter fine-tuning and adaptive learning. Use the region generation network in the Faster R-CNN network architecture to extract useful anchor information. On this basis, perform bounding box regression to improve the target grasping accuracy and obtain the updated weights and pooling parameters of each convolutional layer of the Faster R-CNN network.

[0088] (4b3) Save the obtained data of each layer of the Faster R-CNN network, and use the test set to identify the number of people and whether there is any accidental entry within the perimeter through the Faster R-CNN network trained in step (4b2): When there is a moving target in the restricted area, it indicates that the energized interval has been entered by accident. The edge decision system directly issues an anti-mistake instruction to the anti-mistake host, and the edge decision system will transmit the image data to the cloud decision system for remote control; when the detected target is only one person, it indicates that the operation is performed by a single person. The edge decision system directly issues an anti-mistake instruction to the anti-mistake host, and the edge will transmit the image data to the cloud decision system for remote control.

[0089] The following combination Figure 1 , 2 3. Further explanation of the present invention.

[0090] Power flow analysis is used as the direction for topology generation, and a second round of analysis is employed to prevent missed or over-detected flows. Specifically, the actual power system network is abstracted as a graph, with the positive direction of power flow as the direction of the edges. As can be seen from the process of generating the source-flow path tree, the same flow node may not appear uniquely, often occurring at leaf vertices and branch vertices. To address this characteristic, a second round of topology analysis is performed on the generating source-flow path tree. Based on the distribution of flow nodes in the tree, while generating the complete source-flow path tree, all source-flow path chains originating from the corresponding source are obtained, ultimately yielding the complete set of source-flow path chains for the entire network.

[0091] The division and function of anti-misoperation nodes and anti-misoperation edge zones are mainly based on the line branches of the topology spanning tree as the basis for edge zone division, and the number of devices in the edge zone and the actual area of ​​the line devices as the basis for setting up nodes.

[0092] A cloud-edge collaborative decision-making system was adopted, taking into account actual conditions such as communication and data load, giving the edge decision-making system a certain degree of autonomy, while reducing the pressure on the cloud server.

[0093] Image recognition-based personnel detection and perimeter recognition algorithms were adopted, and the results were used as auxiliary criteria for edge decision-making, thereby making the prevention of misoperation more compliant with regulations.

[0094] The technique of combining Gaussian mixture modeling with Faster R-CNN is adopted. Gaussian mixture modeling is used as a method for target grasping and tracking. The obtained target foreground image is input into the trained Faster R-CNN to obtain the image recognition result and send it to the edge decision system.

[0095] like Figure 1As shown, the intelligent error prevention logic architecture includes modules such as working scenarios, real-time mechanisms, intelligent error prevention models, rule expression, attributes and terminology, rule extraction, and procedural requirements. Rule expression, attributes and terminology, rule extraction, and procedural requirements constitute the process of building the rule base; that is, the output of the rule base needs to be translated into rule expressions. The intelligent error prevention model includes several judgment algorithms, namely, normal operation rule judgment algorithms, special operation data model algorithms, etc., as well as real-time identification of device status. This intelligent error prevention model relies on the real-time mechanism in the working scenario to transmit data. Real-time data acquisition depends on data acquisition devices and video surveillance equipment, including video patrol equipment. The real-time online error prevention mechanism and security protection mechanism correspond to the cloud-edge decision-making module and the cloud-edge image recognition module, respectively.

[0096] In summary, this invention alleviates the problems of excessive server pressure and potential concurrent requests by integrating a cloud-edge collaborative system with image recognition technology, and endows the edge with a certain degree of autonomy, thereby further reducing the reaction time of anti-misoperation judgment; by using the positive power flow as the direction of the edges in graph theory, a complete topology graph is obtained based on two-round topology analysis, and the results are reliable; the anti-misoperation auxiliary verification technology based on image recognition makes the anti-misoperation results more compliant with security specifications.

Claims

1. An intelligent anti-misoperation method based on cloud-edge collaboration and image recognition, characterized in that: The method comprises the following steps in sequence: (1) establishing a rule base: the rule base comprises an electrical rule base and a management rule base; (2) performing power grid topology analysis; (3) constructing a cloud-edge decision system; (4) performing image recognition of the edge decision system, and the image recognition provides auxiliary criteria for decision-making of the edge decision system; The step (2) specifically comprises the following steps: (2a) Constructing a directed graph: abstracting the power network nodes and branches as vertices and edges respectively, and taking the active power flow direction as the direction to obtain the directed graph of the power network , wherein the power grid nodes are abstracted as a vertex set , and the power transmission branch set is abstracted as element arcs , , the starting point of the element arc , and , the ending point of the element arc , thereby obtaining all sources, including incoming lines and buses; (2b) optionally selecting a source as a starting point and marking it as searched; (2c) Establish the adjacency list of the sending node and the power transmission branch, and the vertex set of the directed graph. The set of sending nodes and the set of vertices in the adjacency list of the sending node-power transmission branch. nodes Corresponding to nodes All power transmission branches ; (2d) performing first-round topology analysis: using the established sending end node-sending branch adjacency table, performing topology processing according to active power flow; (2e) performing second-round topology analysis to obtain a complete source-flow path chain; (2f) using a layer-by-layer advancing manner to achieve the purpose of traversal search; (2g) repeating steps (2c) to (2f) until all nodes are traversed; The step (4) specifically comprises the following steps: (4a) Gaussian mixture modeling: (4a1) installing video monitoring equipment at fixed positions and erecting video patrol machines in important areas to obtain clear image data; (4a2) performing background modeling according to the obtained video images, and the pixel time sequence is represented as: ； In the formula, the gray value of the time point , the probability calculation formula of the pixel point at the time point is: ​ ； In the formula, is the moment model weight value; is the moment the mean of the Gaussian distribution; is the moment is the probability density function at the moment, and the calculation formula is: ； (4a3) parameter updating: ； In the formula, is a learning rate, is a threshold coefficient, is moment of time variance of the Gaussian distribution; The Gaussian distribution of the gray value is sorted according to priority, and whether it satisfies formula (1) is detected, if it satisfies, it indicates that the gray value is a matching Gaussian distribution, and the first satisfied value is updated in sequence according to formulas (2) to (5); if it does not satisfy, the data is updated according to formula (6); (4a4) target extraction: After the Gaussian model of video image is established, the Gaussian distribution in the front of the arranged descending order is taken as the background. background:​ ； In the formula, If the current pixel matches the obtained background, the pixel is a moving target. (4b) image recognition: Faster R-CNN network is used as an image recognition tool, and the Faster R-CNN network is trained, and the training process is: (4b1) organizing the extracted targets into a test set, and making a data set obtained after target extraction, and then pasting corresponding labels on the data set; (4b2) inputting the data set into the Faster R-CNN network for parameter fine-tuning, performing self-adaptive learning, using the region generation network of the Faster R-CNN network architecture to extract beneficial anchor information, and performing boundary regression to improve target grabbing accuracy on this basis, to obtain updated weights and pooling parameters of each convolution layer of the Faster R-CNN network; (4b3) saving the obtained Faster R-CNN network layer data, and identifying the number of people and whether there is a misentry in the perimeter in the test set through the Faster R-CNN network trained in step (4b2): when there is a moving target in the forbidden area, it indicates that there is a misentry in the live interval, the edge decision system directly issues a misoperation prevention instruction to the misoperation prevention host, and the edge decision system transmits image data to the cloud decision system for remote control; when the detected target person is only one person, it indicates single-person operation, the edge decision system directly issues a misoperation prevention instruction to the misoperation prevention host, and the edge transmits image data to the cloud decision system for remote control. 2.The intelligent anti-misoperation method based on cloud edge collaboration and image recognition according to claim 1, characterized in that: The step (1) is specifically referring to: by using electrical rules and management rules, the operation rules about the equipment of circuit breaker, disconnector, grounding knife switch, net / cabinet door and its operation mode are combed; the establishment of electrical rule base is according to the substation electrical specification, and the management rule base is established according to the frequently-occurring fault conditions of each substation. 3.The intelligent anti-misoperation method based on cloud edge collaboration and image recognition according to claim 1, characterized in that: The step (3) is specifically referring to: the cloud edge decision system includes a cloud decision system and an edge decision system, and the cloud decision system and the edge decision system both have a cloud edge interaction channel, a rule base, a data acquisition and integration module, a data processing module, an instruction issuing module and a network topology module, the edge decision system further has an image recognition module, and the cloud decision system further has a module for setting the priority of the edge decision system; The cloud decision system and the edge decision system are respectively deployed on the cloud and the on-site edge server, the anti-misoperation host is used as a decision implementation party, the anti-misoperation edge area and the anti-misoperation edge node are used as the data sources of the cloud decision system and the edge decision system, the anti-misoperation host processes the data and images, the anti-misoperation edge area divides the topology spanning tree obtained, takes the line where the substation equipment is located as the division basis, and stores and applies the equipment data and video stream information sent by the anti-misoperation edge node; the video monitoring equipment and the data acquisition equipment are arranged on the anti-misoperation edge node, the video stream images and the equipment state information are collected and sent to the adjacent nodes and the anti-misoperation edge area through wireless transmission; Set one This flag indicates that when communication between the cloud decision system and the edge decision system is normal, this value is 1; otherwise, it is set to 0. When communication is normal, the edge decision system collects data from the error-prevention edge zone and labels each edge zone as follows: ,Will The corresponding data is processed and analyzed, and then sent to the cloud decision-making system; when communication is normal, that is... At that time, the error prevention host receives error prevention instructions from the cloud decision system and sends them to the corresponding devices; when At that time, the side decision-making system directly sends the error prevention information to the error prevention host; When the data to be processed by the cloud decision system exceeds a threshold or the concurrent requests of the edge decision system reach an upper limit, the cloud decision system will frequently send the edge area of the data or the amount of data to the corresponding edge, and the edge decision system will first send the data corresponding to the cloud decision system to the cloud decision system for processing. If the result is a reject operation, the edge will process it by itself, and if the result is an executable operation, the edge will send the result to the anti-misoperation host for simulation and rehearsal. If the result is an executable operation, the cloud will open the corresponding channel when the cloud decision system is idle. At this time, the edge decision system provides data to the cloud again, and the cloud decision system confirms whether the edge lock mutual exclusion is involved.

Images