Push Method, Device and Equipment for Key Points of Safety Risk Control and Inspection in Power Grid On-site Operations

Abstract

The present application discloses a method, device, and equipment for pushing key points of safety risk control and inspection in on-site power grid operations. The method includes: inputting the on-site power grid operation text to be measured into an entity recognition model to identify the target device entity in the on-site operation text, wherein the entity recognition model is trained according to the historical data of on-site power grid operations; matching the safety risk control and inspection key points associated with the target device entity from a preset safety regulation database according to the target device entity; determining the violation management data associated with the safety risk control and inspection key points according to a preset historical violation database; and sorting and pushing the safety risk control and inspection key points to the corresponding terminals according to the violation management data. The present application intelligently pushes the safety risk control and inspection key points and violation management data associated with on-site operations to relevant personnel according to the on-site operation text, ensuring the effectiveness of safety control and inspection and improving the efficiency of safety risk control.

Description

Technical Field

[0001] This application relates to the technical field of power grid safety operation, and in particular to a method, device, and equipment for pushing key points of safety risk control and inspection in on-site power grid operations. Background Art

[0002] There are many construction sites for power grid operations, with a wide range and high risk factors. It is necessary to continuously improve the level of work safety and prevent safety accidents. At present, the safety risk control of on-site operations depends on the one hand on the on-site operators' conscious implementation of the requirements of relevant electric power safety work regulations, and on the other hand on the safety inspectors' analysis of the operators' on-site operation behaviors to determine whether they violate the safety work regulations. For different operation sites and different operation contents, there are a large number of control and inspection key points in the safety work regulations. Manually determining the corresponding control and inspection key points in the safety work regulations for different operation contents is not only inefficient but also unable to ensure that all the control and inspection key points corresponding to the operation content can be obtained. Summary of the Invention

[0003] In view of this, this application provides a method, device, and equipment for pushing key points of safety risk control and inspection in on-site power grid operations. According to the on-site operation text, it intelligently pushes the safety risk control and inspection key points and violation management data associated with the on-site operation to relevant personnel, ensuring the effectiveness of safety control and inspection and improving the efficiency of safety risk control.

[0004] According to one aspect of this application, a method for pushing key points of safety risk control and inspection in on-site power grid operations is provided, including: inputting the on-site power grid operation text to be measured into an entity recognition model to identify the target device entity in the on-site operation text, where the entity recognition model is trained according to the historical data of on-site power grid operations; matching the safety risk control and inspection key points associated with the target device entity from a preset safety regulation database according to the target device entity; determining the violation management data associated with the safety risk control and inspection key points according to a preset historical violation database; and pushing the safety risk control and inspection key points to the terminal corresponding to the on-site operation text according to the violation management data.

[0005] According to another aspect of the present application, there is provided a device for pushing key points of safety risk control and inspection in on-site power grid operations, including: an identification module, configured to input a text of on-site power grid operations to be measured into an entity recognition model, and identify target device entities in the on-site operation text, wherein the entity recognition model is trained based on historical data of on-site power grid operations; a matching module, configured to match, according to the target device entities, key points of safety risk control and inspection associated with the target device entities from a preset safety regulation database; and determine violation management data associated with the key points of safety risk control and inspection according to a preset historical violation database; a pushing module, configured to push the key points of safety risk control and inspection to a terminal corresponding to the on-site operation text according to the violation management data.

[0006] According to still another aspect of the present application, there is provided a computer device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor. When the processor executes the program, the steps of the above-mentioned method for pushing key points of safety risk control and inspection in on-site power grid operations are implemented.

[0007] By means of the above technical solutions, the present application provides a method, a device, and a device for pushing key points of safety risk control and inspection in on-site power grid operations. By using a pre-trained entity recognition model to identify target device entities in the on-site operation text, the target power devices involved in the on-site operation text can be accurately extracted, and key points of safety risk control and inspection associated with the target device entities are matched from a preset safety regulation database, and violation management data associated with the key points of safety risk control and inspection is determined according to a preset historical violation database. Thus, among a large amount of historical violation data and a large number of control and inspection key points included in the safety work regulations, key points of safety risk control and inspection and violation management data related to the target power devices involved in the on-site operation text are accurately extracted, avoiding interference from other irrelevant data to relevant personnel. Furthermore, according to the violation management data, the priority of the key points of safety risk control and inspection is distinguished, and the key points of safety risk control and inspection are pushed to the corresponding terminal, providing accurate reference for relevant personnel, enabling operators to effectively standardize their operation behaviors during the operation according to the pushed data, and enabling inspectors to quickly identify violation behaviors more pertinently according to the pushed data, thereby strengthening the on-site operation safety risk control and ensuring on-site safe operation.

[0008] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the following specifically illustrates the specific embodiments of the present application. Description of the Drawings

[0009] The accompanying drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0010] Figure 1 A flowchart showing the method for pushing key points of safety risk control and inspection in on-site power grid operations provided by an embodiment of the present application is shown;

[0011] Figure 2 A structural block diagram showing the device for pushing key points of safety risk control and inspection in on-site power grid operations provided by an embodiment of the present application is shown;

[0012] Figure 3 A training process diagram for obtaining an entity recognition model provided by an embodiment of the present application is shown. Detailed implementation manners

[0013] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other.

[0014] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application and should not be construed as a limitation to the present application.

[0015] Those skilled in the art of the present technology can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "including" used in the specification of the present application means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say that an element is "connected" or "joined" to another element, it can be directly connected or joined to other elements, or there may also be intermediate elements. In addition, the "connection" or "joining" used herein may include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any unit and all combinations of one or more related listed items.

[0016] Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many different forms and should not be construed as being limited only to the embodiments set forth herein. It should be understood that these embodiments are provided so that the disclosure of the present application is thorough and complete, and the concept of these exemplary embodiments is fully conveyed to those of ordinary skill in the art.

[0017] The method for pushing key points of safety risk control and inspection in on-site power grid operation provided by the embodiments of the present application can be applied to a terminal, or to a server, or can also be software running on a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the method for pushing key points of safety risk control and inspection in on-site power grid operation, etc., but is not limited to the above forms.

[0018] In the production site of the power system, safety risk control and inspection of on-site operations are complex and crucial tasks. Once problems occur in control and inspection, it will seriously affect the safety of on-site operations and may even have a significant adverse impact on the power system. For the specific project construction process, the key points of safety risk control and inspection of on-site operations are different descriptions of the requirements for safety risk control of on-site operations by personnel in different positions. The actual content is basically the same. Therefore, the two can be collectively referred to as control and inspection key points, and the specific content of the control and inspection key points is determined according to the relevant Electric Power Safety Work Regulations of the State Grid Corporation of China (hereinafter referred to as the Safety Regulations). As described in the background technology, at present, the safety risk control of on-site operations mainly relies on manual work. On the one hand, it relies on on-site operation personnel to consciously implement the requirements of relevant safety regulations. On the other hand, it relies on safety inspection personnel to analyze the operation behaviors of operation personnel to determine whether they violate the safety regulations. For different operation contents in different operation scenarios, operation personnel and inspection personnel need to search for the required control and inspection key points in the massive control and inspection key points included in the relevant safety regulations. The workload is huge, and it is impossible to ensure that all the required data can be found, which seriously affects the effectiveness of safety risk control and inspection. In view of this, this application provides a method, device, and equipment for pushing safety risk control and inspection key points of on-site operations in the power grid. According to the text content of the work plan, safety risk control and inspection key points and violation management data associated with the on-site operation are pushed to operation personnel and inspection personnel in advance, which helps operation personnel implement safety risk control and inspection key points during the operation process, effectively helps inspection personnel quickly identify violation behaviors more targeted, and is of great significance for strengthening the safety risk control of the operation site and ensuring the safety of on-site operations.

[0019] In this embodiment, a method for pushing safety risk control and inspection key points of on-site operations in the power grid is provided. As Figure 1 shown, the method includes:

[0020] Step 101, input the text of the on-site operation in the power grid to be measured into the entity recognition model to identify the target device entity in the text of the on-site operation.

[0021] Among them, the entity recognition model is trained according to the historical data of on-site operations in the power grid.

[0022] In this embodiment, the target device entity in the text of the on-site operation is identified through a pre-trained entity recognition model to extract all the target power devices involved in the text of the on-site operation, providing a precise object for subsequent safety risk control and inspection.

[0023] Among them, power devices are devices in the power field, such as transformers, circuit breakers, disconnectors, etc., and other components directly related to the devices.

[0024] In one embodiment, the method for pushing key points of safety risk control and inspection in on-site power grid operations further includes: determining a first training sample set with location tags and a second training sample set without location tags according to historical data of on-site power grid operations; training a preset entity recognition model according to the first training sample set to obtain a first recognition model; inputting the second training sample set into the first recognition model for annotation to determine a third training sample set with predicted tags; performing noise enhancement processing on the third training sample set; training the preset entity recognition model according to the first training sample set and the third training sample set after noise enhancement processing to obtain a second recognition model; and determining the entity recognition model according to the recognition accuracy rate of the first training sample set by the second recognition model.

[0025] Specifically, the entity recognition model is determined according to the preset entity recognition model that has been trained. Among them, the preset entity recognition model is constructed based on a neural network model and is used to identify power equipment entities involved in the input text. Due to the particularity of historical power operation data, it is difficult to obtain a large number of labeled samples. If there are insufficient labeled samples during the training of the preset entity recognition model constructed based on a neural network, the preset entity recognition model will not be able to accurately identify equipment entities in the input text.

[0026] In this embodiment, it is not necessary to annotate all historical data of on-site power grid operations. Only a small amount of historical data is marked as the first training sample set with location tags. The preset entity recognition model is trained according to the first training sample set, and the preset entity recognition model is used to predict the labels of the second training sample set without location tags. The third training sample set with predicted tags is also used as the training data of the preset model to expand the scale of the training data, thereby solving the problem of insufficient labeled samples caused by the difficulty of obtaining a large number of labeled samples in the power operation scenario and ensuring the accuracy of the preset entity recognition model in identifying the input text.

[0027] Specifically, this embodiment trains the preset entity recognition model based on the few-shot learning method of Noisy Student. As Figure 3 shown, a small number of labeled samples are used to train the preset entity recognition model. Then, the preset entity recognition model is used to predict the labels of a large number of unlabeled samples, and noise is added to the samples with predicted labels. Thus, the preset entity recognition model is jointly trained with the labeled samples and the samples with predicted labels after adding noise until the performance of the preset entity recognition model meets the requirements.

[0028] Exemplarily, first, a small amount of historical data of on-site power grid operations is labeled as the first training sample set with location tags according to a preset ratio, and the remaining historical data is determined as the second training sample set without location tags. For example, 1% or less of the historical data is labeled. Then, the preset entity recognition model is trained according to the first training sample set to obtain the first recognition model, and the first recognition model is used as the teacher model.

[0029] Further, the first recognition model is used to label the equipment entities in the second training sample set, generate prediction tags for the second training sample set, and determine the third training sample set with prediction tags. Thus, the labeled training data is determined according to the third training sample set and the first training sample set, the scale of the training data is expanded, and the new preset entity recognition model is trained.

[0030] It is worth mentioning that after obtaining the third training sample set, in this embodiment, strong noise is introduced into the third training sample set by methods such as dropout and random data augmentation, and the new preset entity recognition model is jointly trained according to the first training sample set and the third training sample set after noise augmentation processing, so that the preset entity recognition model can better cope with various noises and uncertainties in the on-site operation text in the power operation scenario, thereby determining the second recognition model, and the second recognition model is used as the student model.

[0031] Further, the first training sample set with location tags is also used as the test set to obtain the recognition accuracy of the second recognition model to verify the recognition performance of the second recognition model for equipment entities. Thus, the entity recognition model is determined according to the recognition accuracy of the second recognition model for the first training sample set.

[0032] Further, as a refinement and extension of the specific implementation manner of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, the entity recognition model is determined according to the recognition accuracy of the second recognition model for the first training sample set, which specifically includes: if the recognition accuracy is less than the second preset threshold, the first recognition model is updated according to the second recognition model until the recognition accuracy is greater than or equal to the second preset threshold; if the recognition accuracy is greater than or equal to the second preset threshold, the second recognition model with the recognition accuracy greater than or equal to the second preset threshold is determined as the entity recognition model.

[0033] In this embodiment, through multiple iterations and self-training of the preset entity recognition model, the recognition performance of the preset entity recognition model for equipment entities is optimized, and the recognition accuracy of the preset entity recognition model for equipment entities is gradually improved.

[0034] Exemplarily, if the recognition accuracy rate is less than the second preset threshold, the first recognition model is updated according to the second recognition model serving as the student model, so that the second recognition model serves as the teacher model, regenerates prediction labels for the second training sample set, and trains a new preset entity recognition model until the recognition accuracy rate of the second recognition model is greater than or equal to the second preset threshold, and the entity recognition model is determined according to the second recognition model with the recognition accuracy rate greater than or equal to the second preset threshold, so that the performance of the entity recognition model in recognizing device entities meets the requirements.

[0035] Among them, the second preset threshold is determined according to the requirements of the actual power operation scenario for the recognition performance of the entity recognition model. For example, the second preset threshold is set to 90%, so that the recognition performance of the entity recognition model for device entities reaches a relatively high level.

[0036] Further, as a refinement and extension of the specific implementation manner of the above embodiment, to fully illustrate the specific implementation process of this embodiment, the steps of training the preset entity recognition model specifically include: constructing structured data in the form of triples according to the preset demand information for extracting power equipment and the training samples input to the preset entity recognition model, where the output data in the structured data is a set of labeled device entities pre-recognized in the input data; based on the BERT model, determining the feature vector matrix of the training samples input to the preset entity recognition model; according to the fully connected layer of the BERT model and the feature vector matrix, determining the entity start position label and the entity end position label of the training samples input to the preset entity recognition model; matching the predicted device entities that conform to the structured data according to the entity start position label and the entity end position label; determining the first component of the loss function corresponding to the entity start position label, the second component of the loss function corresponding to the entity end position label, and the third component of the loss function corresponding to the predicted device entities according to the output data in the structured data; calculating the values of the first component of the loss function, the second component of the loss function, and the third component of the loss function respectively through the cross-entropy function, and performing weighted processing on the calculated values of the first component of the loss function, the second component of the loss function, and the third component of the loss function to determine the target loss function until the target loss function converges.

[0037] In this embodiment, a preset entity recognition model is established based on the BERT model, and structured data in the form of triples is constructed according to the preset demand for extracting power equipment and the training samples input to the preset entity recognition model, where the output data is a set of labeled device entities pre-recognized in the input data, so that the preset entity recognition model can more fully consider the context information in the training samples according to the preset demand for extracting power equipment during the training process, improve the understanding ability of the preset entity recognition model for the context in the training text, and thus ensure the recognition effect of the preset entity recognition model for device entities in different on-site operation texts to adapt to complex power grid operation scenarios.

[0038] Exemplarily, the requirements of the power equipment are specifically described in natural language in advance, and the requirement information Q is determined in the form of a question text. Among them, the requirement information Q is expressed as: Q = {q1, q2, …, q m}, Q is a string, and m is the length of the string.

[0039] Furthermore, according to the requirement information Q, the input data X in the training samples, and the output data y in the training samples, the structural data in the form of a triple (Q, y, X) is constructed. The structural data of the triple is shown in Table 1.

[0040] Among them, using the special characters [CLS] and [SEP] of the BERT model, the requirement information Q is connected with the input data X in the training samples to construct a standard structure, and the standard structure is expressed as

[0041] {[CLS], q1, q2, …, q m , [SEP], x1, x2, …, x n}, X = {x1, x2, …, x n} is a string, and n is the length of the string. The output data y is a set of labeled device entities that have been recognized in advance in the input data X, specifically including the label start position tag x i Begin and the label end position tag x i End of each labeled device entity in the input data X, y = {(x i Begin , x i End ) | i = 1, 2, …, K}, K is the number of labeled device entities included in the input data X. It is worth mentioning that each labeled device entity can be converted into a substring of the input data X.

[0042] Table 1

[0043]

[0044] Furthermore, during the training process, two labels are established for each character in the input data X by the preset entity recognition model, respectively used to judge whether it is the start position or the end position of the device entity.

[0045] Specifically, for an input data X with a length of n in the structural data, the matrix representing its feature vectors is obtained through the BERT model Among them, d is the number of hidden layers. Then, according to the fully connected layer of the BERT model and the feature vector matrix, the entity start position label and the entity end position label of the input data X are determined.

[0046] Exemplarily, the probability matrices P of the entity start position label and the entity end position label in the input data X are obtained through the BERT model Begin and P End are expressed as: P Begin = softmax(M X ·W Begin T ),

[0047] P End = softmax(M X ·W End T ). Among them, are two learnable fully connected layers, which are respectively used to judge the start position and the end position.

[0048] Furthermore, according to the probability matrices P Begin and P End all the entity start position labels of the input data X are determined and the entity end position labels Exemplarily,

[0049] Among them, is the i-th row value of P Begin , corresponding to the i-th character in the input data X, indicates that the i-th character in the input data X is judged as the entity start position by the preset entity recognition model, is the j-th row value of P End , corresponding to the j-th character in the input data X, indicates that the j-th character in the input data X is judged as the entity end position by the preset entity recognition model.

[0050] Furthermore, calculate the start position label and the end position label to match and form the probability value corresponding to the predicted device entity that meets the preset power equipment extraction requirements so as to determine multiple predicted device entities existing in one input data X.

[0051] Among them, is a learnable fully connected layer, and the function concat(·) connects two characters.

[0052] Further, according to the entity start position tag, the entity end position tag, the predicted device entity, and the output data in the structure data, determine the structure of the target loss function of the preset entity recognition model. Calculate each component of the target loss function through the cross-entropy function, and weight each component to obtain the target loss function. Train until the target loss function converges.

[0053] Exemplarily, determine the annotation start position tag x according to the output data y in the training sample. i Begin and the annotation end position tag x i End corresponding probability matrix Y Begin and Y End , and the probability matrix Y corresponding to the annotation device entity Begin,End . Further, according to the probability matrix P corresponding to the start position tag Begin and the probability matrix Y corresponding to the annotation start position tag x i Begin determine the first component of the loss function. According to the probability matrix P Begin corresponding to the end position tag and the probability matrix Y corresponding to the annotation end position tag x End i End determine the second component of the loss function. According to the probability value corresponding to the predicted device entity End Begin,End and the probability matrix Y corresponding to the annotation device entity Begin,End determine the third component of the loss function. Then, determine the target loss function according to the first component, the second component, and the third component of the loss function.

[0054] Specifically, use CE to represent the cross-entropy function, which is used to calculate the values of each component of the loss function. Therefore, the value of the first component of the loss function is expressed as L Begin = CE(P Begin , Y Begin ), the value of the second component of the loss function is expressed as L End = CE(P End , Y End ), the value of the third component of the loss function is expressed as L Span = CE(P Begin,End , Y Begin,End ), and the final target loss function is expressed as L = αL Begin + βL End + γL Span , where α, β, γ ∈ [0, 1] are all preset hyperparameters used to control the proportion of each component of the loss function.

[0055] ​In this embodiment, by constructing structural data including the preset requirements for extracting power equipment, the named entity recognition (NER) task of the BERT model is converted into a machine reading comprehension (MRC) task, so that the trained entity recognition model can better understand the specific context in the power operation scenario of the on-site operation text according to the preset requirements for extracting power equipment, thereby enabling the entity recognition model to better adapt to complex power grid operation scenarios.

[0056] Step 102: Match the safety risk control and inspection key points associated with the target equipment entity from the preset safety regulation database according to the target equipment entity.

[0057] In this embodiment, all the control and inspection key points included in the relevant safety regulations are pre-combed, the source power equipment involved in the control and inspection key points in the safety regulations is extracted, the source equipment entity is determined, the control and inspection key points are sorted out according to the source equipment entity, and a safety regulation database is established. The preset safety regulation database includes the source equipment entity and the control and inspection key points, and there is a corresponding relationship between the source equipment entity and the control and inspection key points.

[0058] It should be noted that in this embodiment, all the control and inspection key points included in the relevant safety regulations are pre-combed, the source power equipment involved in the control and inspection key points in the safety regulations is extracted, the source equipment entity is determined, the control and inspection key points are sorted out according to the source equipment entity, and a safety regulation database is established. The preset safety regulation database includes the source equipment entity and the control and inspection key points, and there is a corresponding relationship between the source equipment entity and the control and inspection key points.

[0059] Further, as a refinement and extension of the specific implementation manner of the above embodiment, to fully illustrate the specific implementation process of this embodiment, step 102, that is, the step of matching the safety risk control and inspection key points associated with the target equipment entity from the preset safety regulation database according to the target equipment entity, specifically includes: matching the target equipment entity with the source equipment entity in the preset safety regulation database, and determining the matching confidence between the target equipment entity and the source equipment entity; if the matching confidence is greater than or equal to the first preset threshold, then determine the control and inspection key points corresponding to the source equipment entity matched with the target equipment entity as the safety risk control and inspection key points associated with the target equipment entity.

[0060] In this embodiment, the target device entity in the on-site operation text is matched with the source device entity in the preset safety regulation database, and the matching degree between the target device entity and the source device entity is quantified by the matching confidence level to prevent incorrect matching, so as to accurately locate the source device entity corresponding to the target device entity in the preset safety regulation database. Furthermore, according to the control and inspection key points corresponding to the source device entity, the safety risk control and inspection key points related to the on-site operation in the relevant safety regulations are accurately extracted, improving the accuracy and comprehensiveness of the information.

[0061] In one embodiment, the method for pushing the safety risk control and inspection key points of the power grid on-site operation further includes: performing text enhancement processing on the source device entity in the preset safety regulation database to obtain a fuzzy device entity; associating the fuzzy device entity with the control and inspection key points corresponding to the source device entity, and updating the source device entity in the preset safety regulation database according to the fuzzy device entity.

[0062] In this embodiment, by performing text enhancement processing on the source device entity, a fuzzy device entity is obtained, so that the fuzzy device entity can cover more device features and descriptions in the power operation scenario to adapt to the diverse expressions of power equipment in the power operation scenario. Thus, the fuzzy device entity is associated with the control and inspection key points corresponding to the source device entity, and the source device entity in the preset safety regulation database is updated according to the fuzzy device entity, enabling the preset safety regulation database to adapt to the newly emerged device description methods, keeping the preset safety regulation database up-to-date, and improving its practicability in different power operation scenarios.

[0063] Exemplarily, the source device entity is obtained from the preset safety regulation database, and text enhancement methods such as back translation are used to generate a fuzzy device entity with a different description but similar meaning to the source device entity. The fuzzy device entity is associated with the control and inspection key points corresponding to the source device entity, and the source device entity in the preset safety regulation database is updated according to the fuzzy device entity. Further, a fuzzy matching model is established based on the BERT model. Specifically, using the word embedding method of the BERT model, the historical device entities in the historical data of the power on-site operation are represented as 768-dimensional high-dimensional feature vectors in the form of sentence vectors through the special character [CLS], and the updated source device entity is used as the label to train the BERT model, so that synonyms are as close as possible in the feature space, thereby associating synonyms with different expressions. Through the above method, each target device entity identified from the on-site operation text can be classified to a source device entity in a preset safety regulation database through the fuzzy matching model. Furthermore, the control and inspection key points corresponding to the source device entity matched with the target device entity are determined as the safety risk control and inspection key points associated with the target device entity.

[0064] Further, to ensure the matching effect between the target device entity and the source device entity, the fuzzy matching model outputs a result only when the matching confidence between the source device entity that matches the target device entity and the target device entity is greater than or equal to the first preset threshold. That is, the fuzzy matching model outputs the source device entity that matches the target device entity only when the matching confidence is greater than or equal to the first preset threshold. This prevents the model from misclassifying the target device entity into a certain source device entity, resulting in incorrect matching. For example, when the target power device in the on-site operation text is not included in the relevant safety regulations, it prevents the blind matching of the fuzzy matching model.

[0065] Exemplarily, where S(x) is the probability distribution confidence (matching confidence) of the target device entity input into the fuzzy matching model, ε is the first preset threshold, and the first preset threshold is determined according to the matching accuracy requirements in the power operation scenario. For example, ε is set to 0.9.

[0066] Step 103: Determine the violation management data associated with the safety risk control and inspection points according to the preset historical violation database.

[0067] It should be noted that, similar to the preset safety regulation database, in this embodiment, the historical violation data of power on-site operations is collected and sorted in advance. The historical violation data is sorted according to the control and inspection points in the preset safety regulation database to determine the number of violations and the risk level corresponding to each control and inspection point, and the number of violations and the risk level are determined as the violation management data, thereby establishing a violation database. The preset historical violation database includes the source device entity and the control and inspection points. And there is a corresponding relationship between the control and inspection points and the violation management data.

[0068] In this embodiment, by determining the violation management data associated with the safety risk control and inspection points of on-site operations, it clarifies the potential risks in the operation process for the operators and avoids the occurrence of violation behaviors. At the same time, it clarifies the risk level corresponding to each operation behavior in the operation process for the inspectors, making the inspection work more targeted. The inspection resources can be reasonably allocated according to the risk level, focusing on the high-risk operation links, and improving the inspection efficiency.

[0069] Exemplarily, the violation management data associated with the control and inspection points corresponding to the safety risk control and inspection points in the preset historical violation database is determined as the violation management data associated with the safety risk control and inspection points.

[0070] Step 104: Push the safety risk control and inspection points to the corresponding terminals according to the violation management data.

[0071] In this embodiment, the priorities of the inspection key points for safety risk control are differentiated according to the violation management data, so as to ensure that the pushed inspection key points for safety risk control are the ones that relevant personnel most hope to obtain, and to ensure the effectiveness of the pushed content.

[0072] Further, as a refinement and extension of the specific implementation manner of the above embodiment, to fully illustrate the specific implementation process of this embodiment, step 104, that is, the step of pushing the inspection key points for safety risk control to the corresponding terminal according to the violation management data, specifically includes: classifying the inspection key points for safety risk control according to preset rules to determine the classification result; sorting the inspection key points for safety risk control according to the violation management data and the classification result to determine the data order; dividing the data order into a first sub-order and a second sub-order according to a preset sequence; pushing the inspection key points for safety risk control in the first sub-order to the terminal corresponding to the on-site operation text in sequence; folding the inspection key points for safety risk control in the second sub-order into a push data group, and pushing the push data group to the terminal corresponding to the on-site operation text.

[0073] In this embodiment, the inspection key points for safety risk control are classified and sorted according to the violation management data to distinguish different inspection key points for safety risk control, obtain the priorities of the inspection key points for safety risk control, prevent the excessive number and content of the inspection key points for safety risk control from affecting the pushing effect, and ensure the effectiveness of the push.

[0074] Specifically, the inspection key points in the preset safety regulation database can be divided into two categories: general inspection key points and special inspection key points. The general inspection key points generally belong to management work, which represents a series of common process operations in the work of safety production control, such as ensuring that construction personnel hold certificates to work, ensuring that the working hours are consistent with the work permit, and the number of workers meets the standard. The special inspection key points are generally related to the operation content, which represents the handling methods for a specific operation in the work of safety production control. For example, for power cables, there are different inspection key points for different operations such as cable head production and cable withstand voltage test.

[0075] Exemplarily, the inspection key points for safety risk control associated with the target equipment entity are divided into special inspection key points for safety risk control and general inspection key points for safety risk control according to the above preset rules. Among them, the inspection key points for safety risk control that meet the preset rules are used as special inspection key points for safety risk control, and the inspection key points for safety risk control that do not meet the preset rules are used as general inspection key points for safety risk control.

[0076] Next, sort the special safety risk control and inspection key points and general safety risk control and inspection key points according to the number of violations and risk levels in the violation management data from high to low to determine the first data order and the second data order. Further, divide the first data order according to the preset order to determine the first sub-order and the second sub-order of the first data order. Then, push the safety risk control and inspection key points in the first sub-order of the first data order to the terminal corresponding to the on-site operation text in sequence to display and push the special control and inspection key points with high risks and high frequencies. Fold the safety risk control and inspection key points in the second sub-order of the first data order into a push data group and push it to the terminal corresponding to the on-site operation text to fold and push the remaining special control and inspection key points with low risks and low frequencies. Similarly, determine the first sub-order and the second sub-order of the second data order according to the preset order, and distinguish the push methods of the general control and inspection key points with different risks and different frequencies according to the above method, which will not be elaborated here.

[0077] It is worth mentioning that relevant personnel of the power grid can all wear mobile communication devices, that is, the present application can correspond to multiple mobile terminals. At this time, the method of the present application is equivalent to a server, which can analyze the on-site operation texts sent by different mobile terminals at the same time and send the safety risk control and inspection key points to the corresponding mobile terminals, so as to meet the complex situation of multiple operators and inspectors working simultaneously in the power operation scenario, improve the efficiency of safety risk control and inspection, and ensure the safety of relevant personnel in the complex power operation scenario. Further, in addition to inputting the on-site operation text in the mobile terminal, the present embodiment can also configure a module for manually selecting operations for the mobile terminal. The module for manually selecting operations includes all the control and inspection key points in the preset safety regulation database. If the user selects one of the control and inspection key points, the module for manually selecting operations will search for the associated source device entity according to the selected control and inspection key point, and use the control and inspection key point corresponding to the source device entity as the safety risk control and inspection key point, thereby increasing the control and inspection methods.

[0078] In one embodiment, the method for pushing the safety risk control and inspection key points for on-site operations in the power grid further includes: if the entity recognition model fails to recognize the target device entity or fails to obtain the safety risk control and inspection key points, push the control and inspection key points that do not meet the preset rules in the preset safety regulation database to the terminal corresponding to the on-site operation text.

[0079] In this embodiment, corresponding push data is determined for special situations to prompt relevant personnel to perform corresponding inspections to meet the complex power grid operation scenarios.

[0080] Exemplarily, if the on-site operation text does not contain power equipment, or the power equipment contained does not belong to the control and inspection objects concerned by the relevant safety regulations, that is, the entity recognition model fails to recognize the target equipment entity, or fails to obtain the control and inspection key points for safety risk control, then the control and inspection key points that do not meet the preset rules in the preset safety regulation database, namely the general control and inspection key points, are determined as the push content. If the relevant personnel receive the push content with only the general control and inspection key points, they can check the work plan to prevent deviations during the operation process.

[0081] This application provides a method, device, and equipment for pushing control and inspection key points for safety risks in on-site power grid operations, which can effectively push the corresponding control and inspection key points for safety risks in relevant safety regulations according to the power equipment involved in the on-site operation text, and is convenient to apply. Compared with the current situation that relies on manual experience summary, the intelligent push method proposed in this application improves work efficiency, reduces the skill requirements for relevant personnel, and can effectively help ensure the safety level of on-site operations.

[0082] In an embodiment, all on-site power operations obtained in a certain province in a certain quarter are used as an application example. The environment used in the application example is: a Core i9-10900K processor with a main frequency of 3.70GHz, a graphics card of NVIDIA GeForce RTX 3090, a video memory size of 24GB, and a memory capacity of 64GB. In the software environment adopted in this article, the Python version is 3.8.10, and the deep learning framework is PyTorch, with its version being 1.11.0. The specific application steps are as follows:

[0083] 1) Construct a historical dataset of on-site power operations. Collect the on-site operation text data of a certain province for one month, a total of 38,293 pieces, which are used to establish an entity recognition model. To verify that this application greatly reduces the requirements for labeled training set data, 200 pieces of data are manually labeled as the first labeled training sample set in the 38,293-piece dataset. At the same time, to verify the effectiveness of the entity recognition model, 4,270 pieces of data are labeled as the test set, and the remaining 33,823 pieces of data form the second unlabeled training sample set.

[0084] 2) Train a preset entity recognition model using the method in the above embodiment of using a small amount of labeled first training sample set and a large amount of unlabeled second training sample set as training data simultaneously. The metrics used to analyze the entity recognition performance are precision, recall, and F1 value. Among them, the F1 value is determined based on the precision and recall. During the training process of the preset entity recognition model, each training completion has experienced a complete 2 epochs, and the iteration process based on the first recognition model and the second recognition model has been carried out for 2 rounds. The device entity recognition results are shown in Table 2. Table 2 shows that if only a small amount of labeled data is used for training, the obtained model will be difficult to achieve the basic function of entity recognition; while after 2 rounds of iteration of the method in this application, the F1 value of device entity recognition has reached 0.9988, indicating that this application can effectively extract device entities in the sample set.

[0085] Table 2

[0086] Iteration round Precision % Recall % F1 value 0 3.95 3.05 0.0344 1 97.48 99.96 0.9870 2 99.78 99.99 0.9988

[0087] 3) When the input text of the on-site operation to be measured is a positive example, that is, the input on-site operation text actually contains power equipment, a complete intelligent push result of safety risk control and inspection key points will be obtained. Take "Remove the branch cable of pole No. * on the 10kV *** line" as a positive example. The target device entity "cable" is identified through the entity recognition model, and the target device entity "cable" is correctly matched with the source device entity "power cable" in the preset safety regulation database. The number of control and inspection key point entries associated with "power cable" obtained through the preset safety regulation database is 28; the number of violations related to the control and inspection key points associated with "power cable" recorded in the preset historical violation database is 40, including 14 special ones and the rest are general. Thus, a push result as shown in Table 3 is obtained.

[0088] Table 3

[0089]

[0090]

[0091] 4) When the input text of the on-site operation to be measured is a negative example, that is, the input on-site operation text does not contain power equipment, or the power equipment contained does not belong to the equipment objects concerned by the relevant safety regulations, general control and inspection key points will be pushed to avoid mis-pushing of special control and inspection key points. Take "Emergency repair operation: Repair of surveillance camera" as a negative example. "Camera" is identified through the entity recognition model. However, the matching confidence of "camera" with the source device entity is lower than 0.9, which means that this target device entity cannot be associated with any source device entity in the preset safety regulation database. Therefore, only general control and inspection key points can be pushed in this example.

[0092] It should be noted that the sequence numbers of the steps in the above embodiments do not indicate the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not impose any limitation on the implementation process of the embodiments of the present application.

[0093] Furthermore, as Figure 2 shown, as a specific implementation of the method for pushing key points of safety risk control and inspection in on-site power grid operations, an embodiment of the present application provides a device 200 for pushing key points of safety risk control and inspection in on-site power grid operations. The device 200 for pushing key points of safety risk control and inspection in on-site power grid operations includes: an identification module 201, a matching module 202, and a pushing module 203.

[0094] Among them, the identification module 201 is configured to input the text of on-site power grid operations to be measured into an entity recognition model to recognize the target device entity in the on-site operation text, where the entity recognition model is trained according to historical data of on-site power grid operations;

[0095] The matching module 202 is configured to match the target device entity with the source device entity in the preset safety regulation database and determine the matching confidence degree between the target device entity and the source device entity; if the matching confidence degree is greater than or equal to a first preset threshold, the control and inspection key points corresponding to the source device entity matched with the target device entity are determined as the safety risk control and inspection key points associated with the target device entity.

[0096] According to the preset historical violation database, determine the violation management data associated with the safety risk control and inspection key points;

[0097] The pushing module 203 is configured to push the safety risk control and inspection key points to the terminal corresponding to the on-site operation text according to the violation management data.

[0098] In one embodiment, the matching module 202 is specifically configured to match the target device entity with the source device entity in the preset safety regulation database and determine the matching confidence degree between the target device entity and the source device entity; if the matching confidence degree is greater than or equal to a first preset threshold, the control and inspection key points corresponding to the source device entity matched with the target device entity are determined as the safety risk control and inspection key points associated with the target device entity.

[0099] The device 200 for pushing key points of safety risk control and inspection in on-site power grid operations further includes:

[0100] An enhancement module, configured to perform text enhancement processing on the source device entity in the preset safety regulation database to obtain a fuzzy device entity; associate the fuzzy device entity with the control and inspection key points corresponding to the source device entity, and update the source device entity in the preset safety regulation database according to the fuzzy device entity.

[0101] In one embodiment, the pushing module 203 is specifically configured to classify the key points for safety risk control and inspection according to a preset rule to determine a classification result; sort the key points for safety risk control and inspection according to the violation management data and the classification result to determine a data order; divide the data order into a first sub-order and a second sub-order according to a preset sequence; sequentially push the key points for safety risk control and inspection in the first sub-order to the terminal corresponding to the on-site operation text; fold the key points for safety risk control and inspection in the second sub-order into a pushed data group, and push the pushed data group to the terminal corresponding to the on-site operation text.

[0102] The pushing device 200 for the key points of safety risk control and inspection in power grid on-site operations further includes:

[0103] A judgment module, configured to, if the entity recognition model fails to recognize the target device entity or fails to obtain the key points for safety risk control and inspection, push the key points for control and inspection in the preset safety regulation database that do not conform to the preset rule to the terminal corresponding to the on-site operation text.

[0104] A training module, configured to determine a first training sample set with location labels and a second training sample set without location labels according to the historical data of power grid on-site operations; train a preset entity recognition model according to the first training sample set to obtain a first recognition model; input the second training sample set into the first recognition model for annotation to determine a third training sample set with predicted labels; perform noise enhancement processing on the third training sample set; train the preset entity recognition model according to the first training sample set and the noise-enhanced third training sample set to obtain a second recognition model; determine the entity recognition model according to the recognition accuracy rate of the first training sample set by the second recognition model.

[0105] In one embodiment, the training module is specifically configured to, if the recognition accuracy rate is less than a second preset threshold, update the first recognition model according to the second recognition model until the recognition accuracy rate is greater than or equal to the second preset threshold; if the recognition accuracy rate is greater than or equal to the second preset threshold, determine the second recognition model with the recognition accuracy rate greater than or equal to the second preset threshold as the entity recognition model.

[0106] In one embodiment, the training module is specifically configured to construct structured data in the form of triples according to the preset demand information for extracting power equipment and the training samples input into the preset entity recognition model, where the output data in the structured data is a set of labeled device entities pre-identified in the input data; determine the feature vector matrix of the training samples input into the preset entity recognition model based on the BERT model; determine the entity start position label and the entity end position label of the training samples input into the preset entity recognition model according to the fully connected layer of the BERT model and the feature vector matrix; match the predicted device entities that conform to the structured data according to the entity start position label and the entity end position label; determine the first component of the loss function corresponding to the entity start position label, the second component of the loss function corresponding to the entity end position label, and the third component of the loss function corresponding to the predicted device entities according to the output data in the structured data; calculate the first component value of the loss function, the second component value of the loss function, and the third component value of the loss function respectively through the cross-entropy function, and perform weighted processing on the calculated first component value of the loss function, the second component value of the loss function, and the third component value of the loss function to determine the target loss function until the target loss function converges.

[0107] For the specific limitations of the push device for the key points of power grid on-site operation safety risk control and inspection, reference can be made to the limitations of the push method for the key points of power grid on-site operation safety risk control and inspection in the above text, which will not be elaborated here. Each module in the above push device for the key points of power grid on-site operation safety risk control and inspection can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor in the computer device in the form of hardware or be independent of it, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0108] Based on the above Figure 1 method shown, and Figure 2 the virtual device embodiment shown, in order to achieve the above purpose, an embodiment of the present application also provides a computer device, which can specifically be a personal computer, a server, a network device, etc. The computer device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the push method for the key points of power grid on-site operation safety risk control and inspection as Figure 1 shown above.

[0109] Optionally, the computer device may further include a user interface, a network interface, a camera, a Radio Frequency (RF) circuit, sensors, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc. Optionally, the user interface may further include a USB interface, a card reader interface, etc. The network interface may optionally include a marked wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0110] Those skilled in the art can understand that the structure of a computer device provided in this embodiment does not constitute a limitation on the computer device, and it may include more or fewer components, or combine certain components, or have different component arrangements.

[0111] The storage medium may further include an operating system and a network communication module. The operating system is a program for managing and storing the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to implement communication between the components inside the storage medium, as well as communication between the storage medium and other hardware and software in the entity device.

[0112] Through the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform, or the embodiments of this application can also be implemented through hardware.

[0113] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the drawings are not necessarily essential for implementing this application. Those skilled in the art can understand that the modules in the devices in the implementation scenario can be distributed in the devices in the implementation scenario according to the description of the implementation scenario, or can be correspondingly changed and located in one or more devices different from this implementation scenario. The modules in the above implementation scenario can be combined into one module, or further split into multiple sub-modules.

[0114] The above serial numbers of this application are only for description and do not represent the advantages or disadvantages of the implementation scenarios. The above disclosure is only several specific implementation scenarios of this application. However, this application is not limited thereto, and any changes that can be thought of by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for pushing key points of safety risk control and inspection in on-site power grid operations, characterized in that, The method includes: Input the on-site operation text of the power grid to be measured into the entity recognition model to recognize the target device entity in the on-site operation text, where the entity recognition model is trained according to the historical data of the on-site operation of the power grid; Match the safety risk control and inspection key points associated with the target device entity from the preset safety regulation database according to the target device entity; Determine the violation management data associated with the safety risk control and inspection key points according to the preset historical violation database; Push the safety risk control and inspection key points to the terminal corresponding to the on-site operation text according to the violation management data; The method further includes: Determine a first training sample set with position labels and a second training sample set without position labels according to the historical data of the on-site operation of the power grid; Train the preset entity recognition model according to the first training sample set to obtain a first recognition model; Input the second training sample set into the first recognition model for annotation to determine a third training sample set with predicted labels; Perform noise enhancement processing on the third training sample set; Train the preset entity recognition model according to the first training sample set and the third training sample set after noise enhancement processing to obtain a second recognition model; Determine the entity recognition model according to the recognition accuracy rate of the first training sample set by the second recognition model; The training of the preset entity recognition model includes: Construct structural data in the form of a triple according to the required information for extracting power equipment preset and the training samples input to the preset entity recognition model, where the output data in the structural data is a set of labeled device entities pre-recognized in the input data; Based on the BERT model, determine the feature vector matrix of the training samples input to the preset entity recognition model; According to the fully connected layer of the BERT model and the feature vector matrix, determine the entity start position label and the entity end position label of the training samples input to the preset entity recognition model; Match the predicted device entity that conforms to the structural data according to the entity start position label and the entity end position label; Determine the first component of the loss function corresponding to the entity start position label, the second component of the loss function corresponding to the entity end position label, and the third component of the loss function corresponding to the predicted device entity according to the output data in the structural data; Calculate the first component value of the loss function, the second component value of the loss function, and the third component value of the loss function respectively through the cross-entropy function, and perform weighted processing on the calculated first component value of the loss function, the second component value of the loss function, and the third component value of the loss function to determine the target loss function until the target loss function converges.

2. The push method for the key points of safety risk control and inspection in on-site power grid operations according to claim 1, characterized in that, The matching of the safety risk control and inspection key points associated with the target device entity from the preset safety regulation database according to the target device entity includes: Match the target device entity with the source device entity in the preset safety regulation database and determine the matching confidence between the target device entity and the source device entity; If the matching confidence is greater than or equal to the first preset threshold, the control and inspection key points corresponding to the source device entity that matches the target device entity are determined as the safety risk control and inspection key points associated with the target device entity.

3. The method for pushing key points of safety risk control and inspection in on-site power grid operations according to claim 2, characterized in that, The method further includes: Performing text enhancement processing on the source device entity in the preset safety regulation database to obtain a fuzzy device entity; Associating the fuzzy device entity with the control and inspection key points corresponding to the source device entity, and updating the source device entity in the preset safety regulation database according to the fuzzy device entity.

4. The method for pushing key points of safety risk control and inspection in on-site power grid operations according to claim 1, wherein The pushing the safety risk control and inspection key points to the terminal corresponding to the on-site operation text according to the violation management data includes: Classifying the safety risk control and inspection key points according to preset rules to determine a classification result; Sorting the safety risk control and inspection key points according to the violation management data and the classification result to determine a data order; Dividing the data order into a first sub-order and a second sub-order according to a preset bit order; Sequentially pushing the safety risk control and inspection key points in the first sub-order to the terminal corresponding to the on-site operation text; Collapsing the safety risk control and inspection key points in the second sub-order into a push data group, and pushing the push data group to the terminal corresponding to the on-site operation text.

5. The pushing method of the inspection key points for the safety risk control of on-site power grid operations according to claim 4, wherein, The method further includes: If the entity recognition model fails to recognize the target device entity or fails to obtain the safety risk control and inspection key points, pushing the control and inspection key points that do not conform to the preset rules in the preset safety regulation database to the terminal corresponding to the on-site operation text.

6. The push method of the key points for the inspection of the safety risk control in on-site power grid operations according to claim 1, wherein, The determining the entity recognition model according to the recognition accuracy rate of the first training sample set by the second recognition model includes: If the recognition accuracy rate is less than the second preset threshold, updating the first recognition model according to the second recognition model until the recognition accuracy rate is greater than or equal to the second preset threshold; If the recognition accuracy rate is greater than or equal to the second preset threshold, determining the second recognition model with the recognition accuracy rate greater than or equal to the second preset threshold as the entity recognition model.

7. A pushing device for inspection key points of safety risk control in on-site power grid operations, characterized in that, The device includes: An identification module, configured to input an on-site power grid operation text to be measured into an entity recognition model to identify a target device entity in the on-site operation text, where the entity recognition model is trained according to historical data of on-site power grid operations; A matching module, configured to match the safety risk control and inspection key points associated with the target device entity from a preset safety regulation database; and determine the violation management data associated with the safety risk control and inspection key points according to a preset historical violation database; A pushing module, configured to push the safety risk control and inspection key points to the terminal corresponding to the on-site operation text according to the violation management data; The pushing device for safety risk control and inspection key points of on-site power grid operations further includes: Determining a first training sample set with location tags and a second training sample set without location tags according to the historical data of on-site power grid operations; Train a preset entity recognition model according to the first training sample set to obtain a first recognition model; Input the second training sample set into the first recognition model for annotation to determine a third training sample set with predicted labels; Perform noise enhancement processing on the third training sample set; Train the preset entity recognition model according to the first training sample set and the third training sample set after noise enhancement processing to obtain a second recognition model; Determine the entity recognition model according to the recognition accuracy of the first training sample set by the second recognition model; The training of the preset entity recognition model includes: Construct structural data in the form of triples according to the demand information for extracting power equipment preset and the training samples input to the preset entity recognition model, where the output data in the structural data is a set of labeled device entities pre-recognized in the input data; Based on the BERT model, determine the feature vector matrix of the training samples input to the preset entity recognition model; According to the fully connected layer of the BERT model and the feature vector matrix, determine the entity start position label and the entity end position label of the training samples input to the preset entity recognition model; Match the predicted device entities that conform to the structural data according to the entity start position label and the entity end position label; Determine the first component of the loss function corresponding to the entity start position label, the second component of the loss function corresponding to the entity end position label, and the third component of the loss function corresponding to the predicted device entity according to the output data in the structural data; Calculate the first component value of the loss function, the second component value of the loss function, and the third component value of the loss function respectively through the cross-entropy function, and perform weighted processing on the calculated first component value of the loss function, the second component value of the loss function, and the third component value of the loss function to determine the target loss function until the target loss function converges.

8. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the program, it implements the method for pushing key points of grid on-site operation safety risk control and inspection as described in any one of claims 1 to 6.

Images