Power distribution network cable discharge hidden danger identification method and system based on cloud edge cooperation

Abstract

The invention discloses a cloud-side collaborative power distribution network cable discharge hidden danger identification method and system. The method comprises the steps of obtaining a severity level, a hidden danger reason and an occurrence position of a hidden danger event; semantic feature extraction is carried out on hidden danger reasons, and hidden danger events are classified into a plurality of risk types; a triple data set composed of severity levels, risk types and environment attributes is obtained; based on the environment attributes of the monitoring area, determining the risk type associated with the monitoring area according to the environment attribute combination strongly associated with the risk type, and determining the weight of the risk type; the edge server configures an identification model according to the risk type in each monitoring area, and configures identification resources for identification tasks corresponding to the risk type according to the weight of each risk type in the monitoring area, so that identification of various discharge hidden dangers is carried out, and the risk type with high weight obtains more identification resources compared with the risk type with low weight.

Description

Technical Field

[0001] The present application relates to edge computing and artificial intelligence, and in particular to a cloud-edge collaborative method and system for identifying potential discharge hazards in distribution network cables. Background Art

[0002] In the related technologies, due to the widespread existence of discharge hazards (also known as hidden danger discharge), such as lightning strikes, icing, tree barriers, pollution flashover, foreign objects hanging, etc., these hazards are common in the distribution network. These risks need to be identified regularly during the operation of the power grid, and the occurrence of these hidden dangers usually has certain environmental characteristics related to the environment in which the distribution network is located. Therefore, some mathematical models or AI models can be used to identify a certain type of discharge hazard. In order to identify different hidden dangers, various types of models have been designed and trained in the prior art. However, there are differences in the types of hidden dangers that occur in different environments. Therefore, if all recognition models are configured for the monitoring area, and all recognition analysis is performed at the same frequency, the efficiency is not high. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides a cloud-edge-coordinated distribution network cable discharge hidden danger identification method and system to optimize the efficiency of identifying cable hidden danger discharge during the operation of the distribution network.

[0004] On the one hand, an embodiment of the present application provides a cloud-edge collaborative distribution network cable discharge hazard identification method, comprising the following steps:

[0005] Obtain the severity level, cause and location of hidden danger events;

[0006] According to the hidden danger causes, hidden danger events are classified into multiple risk types; specifically, semantic features are extracted from hidden danger causes, feature vectors after semantic feature extraction are clustered, clustering results are sampled and analyzed to obtain keywords, and clustering results are screened based on keywords to classify hidden danger events into multiple risk types;

[0007] Based on the map information, the environmental attributes of the location of the occurrence are determined, thereby obtaining a triple data set consisting of severity level-risk type-environmental attribute; through statistical analysis of the triple data set, the combination of environmental attributes that is strongly associated with the risk type is determined;

[0008] Based on the environmental attributes of the monitoring area, determine the risk type associated with the monitoring area according to the combination of environmental attributes strongly associated with the risk type, and determine the weight of the risk type;

[0009] Allocate each monitoring area to an edge server, and one edge server serves multiple monitoring areas;

[0010] The edge server configures the recognition model according to the risk type in each monitoring area, and configures the recognition resources for the recognition task corresponding to the risk type according to the weight of each risk type in the monitoring area, so as to identify various types of discharge hazards. Among them, risk types with high weights are allocated more recognition resources than those with low weights.

[0011] In some embodiments, through statistical analysis of the triplet data set, the combination of environmental attributes strongly associated with the risk type is determined, specifically:

[0012] The environmental attribute combinations involved in the same risk type in the triple data set are statistically analyzed to determine the environmental attribute combinations that are strongly associated with the risk type and that have a number of environmental attributes with a proportion greater than a threshold.

[0013] In some embodiments, based on the environmental attributes of the monitoring area, the risk type associated with the monitoring area is determined according to the combination of environmental attributes strongly associated with the risk type, and the weight of the risk type is determined, specifically:

[0014] The monitoring area is sliced ​​non-overlappingly according to a rectangular sliding window of a certain size to obtain multiple slice units;

[0015] Determine the environmental attributes of the slice unit according to the location of the slice unit, and judge the risk type to which the environmental attributes are adapted;

[0016] According to the risk type adapted by each slice unit, determine all risk types in the monitoring area and the corresponding risk type coverage rate, and determine the weight of the risk type according to the risk type coverage rate and the average severity level corresponding to the risk type;

[0017] The weight of risk type Among them, i, m belong to [1, n], n represents the total number of risk types corresponding to the monitoring area, g i represents the coverage of risk type i, h i Indicates the average severity level corresponding to risk type i.

[0018] In some embodiments, allocating each monitoring area to an edge server specifically includes:

[0019] Construct a risk type vector for each monitoring area. The dimension of the risk type vector is set according to the total number of all risk types. The risk type contained in the monitoring area is set to 1, otherwise it is set to 0. The monitoring areas with the same risk type vector are classified to obtain several subclasses. The modulus of the risk type vector of each subclass is calculated as the vector value. It is determined whether each subclass belongs to another subclass with a larger vector value than itself. If so, the subordinate relationship between the two subclasses is marked.

[0020] Among them, if all risk types of the subclass with a smaller vector value belong to the risk type of the subclass with a larger vector value, the two subclasses have a subordinate relationship; the edge server is allocated to each subclass in the order of the vector value size; when monitoring areas belonging to different subclasses need to be allocated to the same edge server, the subordinate subclasses of the subclass already allocated to the edge server will be allocated first.

[0021] In some embodiments, the monitoring area is sliced ​​non-overlappingly according to a rectangular sliding window of a certain size to obtain a plurality of slice units, specifically:

[0022] Select the size of the slice unit according to the size of the monitoring area;

[0023] Map the monitoring area into the coordinate system, and map the segmentation matrix into the coordinate system according to the size of the slice unit so that the segmentation matrix overlaps with the monitoring area;

[0024] The units in the segmentation matrix that are surrounded by the monitoring area are determined as slice units.

[0025] In some embodiments, the following steps are also included:

[0026] Based on the user's request on the browser or client, the map is loaded, and the risk information of each area is marked on the map based on the model prediction results reported by the edge server.

[0027] In some embodiments, the hidden danger cause is a result of a text description, and extracting semantic features from the hidden danger cause and clustering the feature vector after the semantic feature extraction specifically includes:

[0028] The semantic features of hidden danger causes are extracted through the trained general semantic model and mapped into N-dimensional feature vectors;

[0029] The characteristic vectors of multiple hidden danger causes are extracted as the initialization center, the center described by each sample is traversed and the center position is recalculated, and the required number of iterations is performed to obtain the final clustering result.

[0030] In some embodiments, the model deployed on the edge server is uniformly trained by the cloud and then sent down to be deployed in the edge server.

[0031] On the other hand, an embodiment of the present application provides a cloud-edge collaborative distribution network cable discharge hazard identification system, including:

[0032] Memory, used to store programs;

[0033] A processor is used to load the program to execute the cloud-edge collaborative distribution network cable discharge hazard identification method.

[0034] On the other hand, an embodiment of the present application provides a cloud-edge collaborative distribution network cable discharge hazard identification system, including a cloud and edge server;

[0035] The cloud is used to: obtain the severity level, cause of the hidden danger and location of the hidden danger event; extract semantic features of the hidden danger cause, cluster the feature vectors after the semantic feature extraction, sample and analyze the clustering results to obtain keywords, and screen the clustering results based on the keywords to classify the hidden danger events into multiple risk types; determine the environmental attributes of the location of the occurrence based on map information, so as to obtain a triple data set consisting of severity level-risk type-environmental attributes; determine the combination of environmental attributes strongly associated with the risk type through statistical analysis of the triple data set; based on the environmental attributes of the monitoring area, determine the risk type associated with the monitoring area according to the combination of environmental attributes strongly associated with the risk type, and determine the weight of the risk type; allocate each monitoring area to an edge server, and one edge server serves multiple monitoring areas;

[0036] The edge server is used to configure the recognition model according to the risk type in each monitoring area, and configure the recognition resources for the recognition task corresponding to the risk type according to the weight of each risk type in the monitoring area, so as to identify various hidden dangers. Among them, risk types with high weights are allocated more recognition resources than those with low weights.

[0037] Beneficial effects: This solution performs semantic analysis on the severity level, cause of hidden dangers and location of hidden danger events in historical data, so as to classify the data into multiple risk types that can be identified by the model. Then, the environmental attributes of the location of occurrence are determined based on the map information, so as to obtain a triple data set consisting of severity level-risk type-environmental attributes, and then the weights of various risk types in the monitoring area are analyzed based on the triples. Then, the edge server configures the recognition model according to the risk type in each monitoring area, and configures the recognition resources for the recognition tasks corresponding to the risk type according to the weight of each risk type in the monitoring area, so as to identify various types of discharge hidden dangers, among which, risk types with high weights receive more recognition resources than those with low weights. Through this solution, resources can be optimized according to local conditions, and more resources can be allocated to types with higher risks, so that they can be monitored more frequently. This solution optimizes the recognition efficiency of cable hidden danger discharge during the operation of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments.

[0039] Figure 1 is a flow chart of a method provided by an embodiment of the present application;

[0040] Figure 2 It is a system block diagram provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described through the implementation method with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0042] Reference Figure 1 and Figure 2 , the embodiment of the present application provides a cloud-edge collaborative hidden danger identification method, such as Figure 2 As shown, in the present application, the system structure includes a cloud and an edge server, and the cloud and the edge server can communicate through the Internet of Things or the Internet. The cloud can configure relevant policies for the edge server. As can be seen from the figure, an edge server can monitor multiple areas, and for different areas, different tasks can be monitored, or the same task can be monitored. For different areas, different sizes of resources can be allocated to different tasks. Among them, the size of the task in the figure represents its allocated resources. It can be understood that for tasks with higher weights, more resources can be configured, and these resources support tasks to use higher performance solutions or to be executed at a higher frequency.

[0043] The following steps are involved:

[0044] S1. Obtain the severity level, cause of the hidden danger and location of the hidden danger event. It is understandable that through the analysis of historical reports, the severity level of each event and the description of the cause of the hidden danger can be obtained, where the cause description is equivalent to a text description of the nature of a briefing, and the location of the occurrence can be described by map location. Usually, when checking such information, a corresponding report will be formed, and the characteristics of the hidden danger can be obtained by analyzing the historical reports. It is understandable that there are various situations for hidden danger events, but it may not be possible to directly classify the data by type class in historical data.

[0045] S2. Extract semantic features from hidden danger causes, cluster feature vectors after semantic feature extraction, sample and analyze clustering results to obtain keywords, and filter clustering results based on keywords to classify hidden danger events into multiple risk types.

[0046] It is understandable that since the hidden danger causes are described in text in a non-standardized way, data structure processing is required. In general, the sentence structure and style of the report description are basically the same, and the content will contain specific causes of the incident or hidden dangers. These causes are classified into certain risk types. Therefore, the text is extracted into a vector representation in the semantic space through a general semantic model. It is understandable that these semantic models can be BERT models, GPT models, paragraph2vec or doc2vec, etc. In this step, the text is mapped to a vector representation in the model semantic space through the NLP model. Based on this, the content can be clustered based on the vector representation.

[0047] Then after clustering, the actual keywords involved in the clustering are determined by sampling analysis. This step can find the words used to express the relevant risk types in the causes. After keyword screening, samples of the corresponding causes can be filtered out, so that the clustering algorithm can avoid classifying different risk types with similar expressions into one family. In this process, since the keywords used in the accident causes of events belonging to a certain risk type cannot be determined in advance, it is necessary to perform certain sampling after clustering to determine the keywords. It is understandable that by classifying the samples corresponding to the risk types, it is helpful to further analyze the environmental characteristics of the risk types. It is understandable that the risk type is a type corresponding to at least one risk prediction model. It is understandable that the risk prediction model described in this application is an existing model. This application does not improve these models, but only discusses how to allocate resources for these diverse models. These models can be models for predicting the probability of hidden danger discharge caused by tree barriers, or models for predicting the degree of danger of hidden danger discharge caused by rain. This application attempts to allocate resources corresponding to different risk types by analyzing the relationship between the environmental characteristics of the area and the hidden danger discharge type.

[0048] In step S2, semantic features are extracted from the hidden danger causes, and feature vectors after semantic feature extraction are clustered, specifically including:

[0049] S21. The semantic features of the hidden danger causes are extracted through a trained general semantic model and mapped into an N-dimensional feature vector. It is understandable that the NLP model trained in Chinese can map the input sentences and paragraphs into the semantic space and output a semantic vector. It is understandable that the vector-based representation of the model is actually the weighted superposition of the sum of the meanings and order relationships of each word segment. Therefore, the vector output by the model represents the overall meaning, and the length of the vector can output a vector result of the same length regardless of the length difference of the input.

[0050] S22, extract multiple characteristic vectors of hidden danger causes as initialization centers, traverse the centers described by each sample and recalculate the center position, iterate the required number of times, and obtain the final clustering result. This solution can use the kmean clustering algorithm, first initialize K centers, then cluster the vectors by calculating the distance between the vectors, and update the center position. Then perform iterative optimization until the number of iterations is met.

[0051] S3. Based on the map information, determine the environmental attributes of the location of occurrence, thereby obtaining a triple data set consisting of severity level-risk type-environmental attributes; through statistical analysis of the triple data set, determine the combination of environmental attributes that is strongly associated with the risk type.

[0052] Specifically, through statistical analysis of the triplet data set, the combination of environmental attributes that are strongly associated with the risk type is determined:

[0053] S31. Count the combinations of environmental attributes involved in the same risk type in the triple data set, and determine that a number of environmental attributes with an appearance ratio greater than a threshold constitute a combination of environmental attributes strongly associated with the risk type.

[0054] It is understandable that by counting the occurrence ratio of the same risk type, the environmental attributes of a certain proportion are obtained, that is, the proportion of environmental attributes in the total sample. One solution is to directly determine the combination of these environmental attributes as the environmental attribute combination. Another solution is to perform a secondary screening according to the proportion of the combination. After combining the environmental attributes, the combinations with a proportion greater than the threshold are screened out as strongly correlated environmental attribute combinations. It is understandable that a strong correlation means that there is a close relationship between the two. That is, the occurrence of one party has a relatively large relationship with the other party. It is understandable that this step actually determines the environmental attribute combination according to a certain algorithm.

[0055] S4. Based on the environmental attributes of the monitoring area, determine the risk type associated with the monitoring area according to the combination of environmental attributes strongly associated with the risk type, and determine the weight of the risk type; it can be understood that environmental attributes include building type, road type, river type, line, green type, climate type, etc.

[0056] Step S4 specifically includes:

[0057] S41, slicing the monitoring area into non-overlapping slices according to a rectangular sliding window of a certain size to obtain a plurality of slice units.

[0058] Specifically, the size of the slice unit is selected according to the size of the monitoring area; the monitoring area is mapped into the coordinate system, and the segmentation matrix is ​​mapped into the coordinate system according to the size of the slice unit so that the segmentation matrix overlaps with the monitoring area; and the units in the segmentation matrix surrounded by the monitoring area are determined as slice units.

[0059] It is understandable that for a larger monitoring area, the environmental elements it contains may be diverse. However, for a small area, this may not be the case. The environmental elements required for risk need to be concentrated in a small area. Therefore, in this scheme, a large area is sliced ​​to analyze the actual environmental conditions of the slice units. As for the monitoring area, it is not demarcated as a regular area. During analysis, in order to facilitate segmentation, it is segmented by regular shapes. Therefore, the monitoring area can be mapped to the coordinate system and then segmented in the form of a rectangular sliding window.

[0060] S42: Determine the environmental attributes of the slice unit according to the location of the slice unit, and judge the risk type to which the environmental attributes are adapted. It is understandable that the environmental attributes of each slice unit can be analyzed, and then the risk type that may exist in the environmental attributes can be matched.

[0061] S43, determining all risk types and corresponding risk type coverage rates of the monitoring area according to the risk types adapted by each slice unit, and determining the weights of the risk types according to the risk type coverage rates and the average severity levels corresponding to the risk types;

[0062] The weight of risk type Among them, i, m are positive integers belonging to [1, n], n represents the total number of risk types corresponding to the monitoring area, g i represents the coverage of risk type i, h i Indicates the average severity level corresponding to risk type i.

[0063] It can be understood that, corresponding to a number of environmental attributes, as long as the combination of environmental attributes corresponding to the risk type is met, it can be considered that the corresponding risk type may exist in the environment. Similarly, the aforementioned can calculate the severity level of the corresponding risk type in the history for the same risk type, thereby calculating the weight of the risk type through the above weight calculation method.

[0064] S5. Allocate each monitoring area to an edge server, and one edge server serves multiple monitoring areas. It can be understood that the monitoring areas are allocated to multiple edge servers to perform regular predictions of corresponding risk models.

[0065] Specifically, construct a risk type vector for each monitoring area, and set the dimension of the risk type vector according to the total number of all risk types; set the risk type contained in the monitoring area to 1, otherwise set it to 0, classify the monitoring areas with the same risk type vector to obtain several subclasses, calculate the modulus of the risk type vector of each subclass as the vector value, and judge whether each subclass belongs to another subclass with a larger vector value than itself. If so, mark the subordinate relationship between the two subclasses;

[0066] Among them, if all risk types of the subclass with a smaller vector value belong to the risk type of the subclass with a larger vector value, the two subclasses have a subordinate relationship; the edge server is allocated to each subclass in the order of the vector value size; when monitoring areas belonging to different subclasses need to be allocated to the same edge server, the subordinate subclasses of the subclass already allocated to the edge server will be allocated first.

[0067] Based on the classification results, monitoring tasks are assigned to edge servers in the edge server group. It is understandable that the edge server only needs to be assigned to monitor all risk types in its corresponding monitoring area. Therefore, if the risk types of the monitoring areas are the same, the number of models configured by the edge server can be reduced. For example, the risks include four categories: A, B, C, and D. The monitoring areas with only risks A and B can be centrally assigned to one edge server. In this way, the server does not need to identify C and D.

[0068] S6. The edge server configures the identification model according to the risk type in each monitoring area, and configures the identification resources for the identification task corresponding to the risk type according to the weight of each risk type in the monitoring area, wherein the risk type with a high weight gets more identification resources than the risk type with a low weight. It can be understood that several risk identification models will be run for each monitoring area, and resources can be allocated according to the weight of each model corresponding to the area. Those with high weights will obtain higher identification frequencies or higher identification priorities. Specific strategies can be adjusted according to the situation.

[0069] S7. According to the user's request on the browser or client, the map is loaded, and the risk information of each area is marked on the map according to the model prediction results reported by the edge server. It can be understood that the user can log in through the client and display the model prediction results by area in the form of a visual map, and conduct inspections and troubleshooting based on the above information.

[0070] In some embodiments, the model deployed on the edge server is uniformly trained by the cloud and then sent down to be deployed in the edge server.

[0071] A cloud-edge collaborative distribution network cable discharge hazard identification system, comprising:

[0072] Memory, used to store programs;

[0073] A processor is used to load the program to execute a cloud-edge collaborative distribution network cable discharge hidden danger identification method. A cloud-edge collaborative distribution network cable discharge hidden danger identification system includes a cloud and an edge server;

[0074] The cloud is used to: obtain the severity level, cause of hidden danger and location of the accident; extract semantic features of the cause of hidden danger, cluster the feature vectors after semantic feature extraction, sample and analyze the clustering results to obtain keywords, and filter the clustering results based on the keywords to classify hidden danger events into multiple risk types; determine the environmental attributes of the location of occurrence based on map information, so as to obtain a triple data set consisting of severity level-risk type-environmental attributes; determine the combination of environmental attributes strongly associated with the risk type through statistical analysis of the triple data set; based on the environmental attributes of the monitoring area, determine the risk type associated with the monitoring area according to the combination of environmental attributes strongly associated with the risk type, and determine the weight of the risk type; assign each monitoring area to an edge server, and one edge server serves multiple monitoring areas;

[0075] The edge server is used to configure the recognition model according to the risk type in each monitoring area, and to configure the recognition resources for the recognition tasks corresponding to the risk type according to the weight of each risk type in the monitoring area, wherein risk types with high weights are allocated more recognition resources than those with low weights.

[0076] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0077] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0078] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0079] Note that the above are only preferred embodiments of the present application and the technical principles used. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A cloud-edge collaborative distribution network cable discharge hazard identification method, characterized in that: The following steps are involved: Obtain the severity level, cause and location of hidden danger events; According to the causes of hidden dangers, hidden danger events are classified into multiple risk types; Based on the map information, the environmental attributes of the location of the occurrence are determined, thereby obtaining a triple data set consisting of severity level-risk type-environmental attribute; through statistical analysis of the triple data set, the combination of environmental attributes that is strongly associated with the risk type is determined; Based on the environmental attributes of the monitoring area, determine the risk type associated with the monitoring area according to the combination of environmental attributes strongly associated with the risk type, and determine the weight of the risk type; Allocate each monitoring area to an edge server, and one edge server serves multiple monitoring areas; The edge server configures the recognition model according to the risk type in each monitoring area, and configures the recognition resources for the recognition task corresponding to the risk type according to the weight of each risk type in the monitoring area, so as to identify various types of discharge hazards. Among them, risk types with high weights are allocated more recognition resources than those with low weights.

2. The method according to claim 1, characterized in that Through statistical analysis of the triplet data set, the combination of environmental attributes strongly associated with the risk type is determined, specifically: The environmental attribute combinations involved in the same risk type in the triple data set are statistically analyzed to determine the environmental attribute combinations that are strongly associated with the risk type and that have a number of environmental attributes with a proportion greater than a threshold.

3. The method according to claim 1, characterized in that: Based on the environmental attributes of the monitoring area, the risk type associated with the monitoring area is determined according to the combination of environmental attributes strongly associated with the risk type, and the weight of the risk type is determined, specifically: The monitoring area is sliced ​​non-overlappingly according to a rectangular sliding window of a certain size to obtain multiple slice units; Determine the environmental attributes of the slice unit according to the location of the slice unit, and judge the risk type to which the environmental attributes are adapted; According to the risk type adapted by each slice unit, determine all risk types in the monitoring area and the corresponding risk type coverage rate, and determine the weight of the risk type according to the risk type coverage rate and the average severity level corresponding to the risk type; The weight of risk type Among them, i, m are positive integers belonging to [1, n], n represents the total number of risk types corresponding to the monitoring area, g i represents the coverage of risk type i, h i Indicates the average severity level corresponding to risk type i.

4. The method according to claim 3, characterized in that The allocating each monitoring area to the edge server specifically includes: Construct a risk type vector for each monitoring area. The dimension of the risk type vector is set according to the total number of all risk types. The risk type contained in the monitoring area is set to 1, otherwise it is set to 0. The monitoring areas with the same risk type vector are classified to obtain several subclasses. The modulus of the risk type vector of each subclass is calculated as the vector value. It is determined whether each subclass belongs to another subclass with a larger vector value than itself. If so, the subordinate relationship between the two subclasses is marked. Among them, if all risk types of the subclass with a smaller vector value belong to the risk type of the subclass with a larger vector value, the two subclasses have a subordinate relationship; the edge server is allocated to each subclass in the order of the vector value size; when monitoring areas belonging to different subclasses need to be allocated to the same edge server, the subordinate subclasses of the subclass already allocated to the edge server will be allocated first.

5. The method according to claim 1, characterized in that The monitoring area is sliced ​​non-overlappingly according to a rectangular sliding window of a certain size to obtain multiple slice units, specifically: Select the size of the slice unit according to the size of the monitoring area; Map the monitoring area into the coordinate system, and map the segmentation matrix into the coordinate system according to the size of the slice unit so that the segmentation matrix overlaps with the monitoring area; The units in the segmentation matrix that are surrounded by the monitoring area are determined as slice units.

6. The method according to claim 1, characterized in that According to the causes of hidden dangers, hidden danger events are classified into multiple risk types, including: Semantic features are extracted from the causes of hidden dangers, and the feature vectors after semantic feature extraction are clustered. The clustering results are sampled and analyzed to obtain keywords, and the clustering results are screened based on the keywords to classify the hidden danger events into multiple risk types.

7. The method according to any one of claims 1 to 6, characterized in that: The hidden danger cause is the result of text description, and the semantic feature extraction of the hidden danger cause and clustering of the feature vector after the semantic feature extraction specifically include: The semantic features of hidden danger causes are extracted through the trained general semantic model and mapped into N-dimensional feature vectors; The characteristic vectors of multiple hidden danger causes are extracted as the initialization center, the center described by each sample is traversed and the center position is recalculated, and the required number of iterations is performed to obtain the final clustering result.

8. The method according to any one of claims 1 to 6, characterized in that: The models deployed on the edge servers are trained uniformly in the cloud and then deployed on the edge servers.

9. A cloud-edge collaborative distribution network cable discharge hazard identification system, characterized in that: include: Memory, used to store programs; A processor is used to load the program to execute the cloud-edge collaborative distribution network cable discharge hazard identification method as described in any one of claims 1-8.

10. A cloud-edge collaborative distribution network cable discharge hazard identification system, characterized in that: Includes cloud and edge servers; The cloud is used to: obtain the severity level, cause and location of hidden danger events; and classify hidden danger events into multiple risk types according to the cause of hidden dangers; Based on the map information, the environmental attributes of the location of the occurrence are determined, thereby obtaining a triple data set consisting of severity level-risk type-environmental attribute; through statistical analysis of the triple data set, the combination of environmental attributes strongly associated with the risk type is determined; based on the environmental attributes of the monitoring area, the risk type associated with the monitoring area is determined according to the combination of environmental attributes strongly associated with the risk type, and the weight of the risk type is determined; each monitoring area is assigned to an edge server, and one edge server serves multiple monitoring areas; The edge server is used to configure the recognition model according to the risk type in each monitoring area, and configure the recognition resources for the recognition task corresponding to the risk type according to the weight of each risk type in the monitoring area, so as to identify various hidden dangers. Among them, risk types with high weights are allocated more recognition resources than those with low weights.

Images