Power grid monitoring risk early warning pushing method and system

Abstract

The invention provides a power grid monitoring risk early warning pushing method and system, and relates to the technical field of power grid monitoring risk early warning, and the method comprises the steps: obtaining a risk early warning description text of a power grid monitoring system, employing a network nerve pre-training model, combining with a keyword dictionary and a regular matching rule, and extracting equipment entity information in the text; in combination with subject attribute data, basic rule screening, machine learning optimization matching and conflict verification correction are performed on equipment entity information, and an early warning notification subject list is output; determining a target push channel combination corresponding to the list through channel fitness calculation, channel state verification and push execution and feedback in combination with channel attribute categories and main body channel preferences; and pushing priority ranking is performed on the list to generate an early warning notification main body pushing list, and risk early warning information is pushed to each early warning notification main body in combination with the target pushing channel combination, so that the accuracy of the early warning pushing main body and the real-time performance of notification are improved, and the power grid risk disposal efficiency is optimized.

Description

Technical Field

[0001] This invention relates to the field of power grid monitoring risk early warning technology, and in particular to a method and system for power grid monitoring risk early warning push. Background Technology

[0002] Traditional power grid monitoring risk warning push methods have advantages such as full-process automation, multi-channel combination, and strong scenario adaptability, which can shorten the warning delivery time and improve the coverage. However, they have obvious defects in practical applications: when faced with new warning descriptions such as new equipment failures and rare weather impacts, pre-trained models have difficulty accurately extracting equipment entities. They also have poor adaptability to low-frequency, high-risk, small-sample scenarios and are prone to omissions. At the same time, this method is highly dependent on data quality. Incomplete equipment ledgers and lack of historical collaborative data for newly added departments can lead to deviations in entity extraction and inaccurate matching calculations, directly affecting the subject matching and channel selection effects, making it difficult to meet the power grid's accurate warning needs.

[0003] Therefore, it is necessary to provide a method and system for power grid monitoring risk early warning to solve the above-mentioned technical problems. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a power grid monitoring risk early warning push method and system, which solves the problems of inaccurate matching, low efficiency, unreliable reach, and ambiguous responsibility in traditional power grid monitoring risk early warning push methods, thus failing to achieve intelligent, automated, and precise early warning notifications.

[0005] This invention provides a method for pushing out power grid monitoring risk warnings, the method comprising: The risk warning description text generated by the power grid monitoring system is obtained, and the equipment entity information in the risk warning description text is extracted by using a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules. Based on the subject attribute data in the subject information database, the device entity information is sequentially subjected to basic rule filtering, machine learning optimization matching, and conflict verification and correction, and a list of warning notification subjects is output. Based on the aforementioned list of entities for early warning notifications, and in conjunction with channel attribute categories and entity channel preferences, the target push channel combination corresponding to the aforementioned list of entities for early warning notifications is determined through channel adaptability calculation, channel status verification, and push execution and feedback. The list of entities receiving early warning notifications is sorted by push priority to generate an early warning notification subject push list. Combined with the target push channel combination, risk warning information is pushed to each entity in the early warning notification subject push list in sequence. The reception status of each entity receiving early warning notifications is tracked in real time, and timeout re-push and upgrade handling are performed.

[0006] Preferably, the step of using a pre-trained neural network model combined with a keyword dictionary and regular expression matching rules to extract device entity information from the risk warning description text specifically includes: The risk warning description text is subjected to noise reduction, word segmentation and standardization processing to unify the abbreviation and colloquial expression format of the device name in the risk warning description text; Obtain equipment entity annotation data in the power grid field and input it into the network neural pre-training model. Perform domain adaptation fine-tuning on the network neural pre-training model, and process the risk warning description text based on the network neural pre-training model to output entity recognition results. A preset keyword dictionary and regular expression matching rules are used to supplement and verify the entity recognition results, and the regular expression matching results are output. The entity recognition result and the regular expression matching result are fused to form structured device entity information, wherein the device entity information includes device type, device name, device voltage level, device number, device region, and device importance.

[0007] Preferably, the equipment type T, equipment voltage level V, equipment region R, and equipment importance S are extracted from the equipment entity information, and the risk level L of the warning-marked equipment is obtained. The risk level L of the warning-marked equipment is divided into 4 levels, and the higher the risk level L of the warning-marked equipment, the greater the risk of the warning-marked equipment. Retrieve the subject attribute data of all subjects from the subject information database, wherein the subject attribute data includes subject department type D and subject jurisdiction area. Scope of Main Responsibilities The frequency of historical collaboration among entities is Y, and all entities in the entity information database are set as follows: m is the total number of entities in the entity information database; For the aforementioned main jurisdiction area The scope of responsibilities of the main body The encoding conversion is performed, and the historical coordination frequency Y is normalized and preprocessed. Establish the association dimension between the device entity information and the main attribute data to form a dataset to be screened.

[0008] Preferably, the dataset to be filtered is filtered using basic rules, specifically including: A pre-defined hard rule base for subject matching is provided, and the hard rule base includes 3 hard rules; Hard rule 1 is applied to the dataset to be filtered to determine the jurisdiction of each subject. Whether the region R to which the device belongs is included, i.e., whether the condition is met. Filter subjects that meet the hard rule 1; For entities that meet the requirements of hard rule 1, hard rule 2 is executed to determine the scope of responsibility of each entity. Whether the device type T is included, i.e., whether the condition is met. Filter subjects that meet the hard rule 2 and add them to the candidate subject set. ; Hard rule 3 is that when the risk level L of the warning labeling device is ≥3, the superior competent authority shall be directly added to the candidate entity set. .

[0009] Preferably, for the candidate subject set Performing machine learning to optimize matching specifically includes: Five feature variables were defined, including the jurisdictional matching degree between the subject and the device. The correlation between the main responsibilities and the types of early warnings Risk level normalized value Normalized value of historical collaboration frequency Response time normalized value ; Among them, the jurisdiction matching degree between the subject and the equipment If the candidate subject set If the candidate subject and the device are a perfect match, then If the candidate subject partially matches the device, then If the candidate entity and the device are indirectly matched, then ; Regarding the correlation between the main responsibilities and the warning types If the primary responsibility of the candidate entity is directly related to the type of early warning, then If the primary responsibility of the candidate entity and the type of early warning are related, then If the primary responsibility of the candidate entity and the type of early warning are related as auxiliary responsibilities, then ; For the risk level normalized value If the risk level L of the warning labeling device is level 1, then If the risk level L of the warning labeling device is level 2, then If the risk level L of the warning labeling device is level 3, then If the risk level L of the warning labeling device is level 4, then ; For any candidate subject The normalized value of the historical collaboration frequency That is, the candidate subject Number of similar warnings processed in history The normalized value, corresponding to the calculation formula is: n is the set of candidate subjects The total number of candidate entities; For the normalized value of response time If the historical average response time of the candidate entity is less than 5 minutes, then If the historical average response time of the candidate entity is 5 to 15 minutes, then If the historical average response time of the candidate entity is greater than 15 minutes, then ; The matching degree between the candidate subjects and this early warning is calculated using a logistic regression algorithm. Among them, the matching degree weight ; In the candidate subject set In this process, candidate entities with a matching score greater than or equal to 0.6 are first added to the list of entities to be verified. Then, the matching scores of other candidate entities are sorted from largest to smallest. The candidate entities corresponding to the top N matching scores are selected and added to the list of entities to be verified. The value of N is dynamically adjusted according to the risk level L of the warning labeling device.

[0010] Preferably, conflict verification and correction are performed on the list of entities to be verified, and the list of entities to receive the early warning notification is output, specifically including: Perform hierarchical verification on the entities to be verified in the list of entities to be verified. If entities to be verified at different levels in the same department are matched repeatedly, only the entities at the highest level are retained, and a list of deduplicated entities is output. Determine if the deduplication subject list is empty. If it is, perform rule downgrading on the hard rule base and re-execute basic rule matching and machine learning optimization matching until the deduplication subject list is not empty, then output the warning notification subject list. .

[0011] Preferably, a set of notification channels is constructed. ,in, Indicates the scheduling screen, Indicates mobile app, Indicates text message, Indicates telephone, Indicates the emergency command system; Combined with the aforementioned channel attribute categories, including delivery timeliness Accessibility Interference Define the attributes of each notification channel in the notification channel set, wherein, for the delivery timeliness... The delivery time for the dispatch screen is 5 seconds, the delivery time for the mobile APP is 3 seconds, the delivery time for the SMS is 10 seconds, the delivery time for the phone call is 1 second, and the delivery time for the emergency command system is 8 seconds; regarding the reachability... The reachability of the dispatch screen is 90%, the reachability of the mobile APP is 95%, the reachability of SMS is 98%, the reachability of telephone is 99%, and the reachability of the emergency command system is 85%; regarding the interference level... The interference level of the dispatch screen is 0.3, the interference level of the mobile APP is 0.6, the interference level of the SMS is 0.7, the interference level of the telephone is 0.9, and the interference level of the emergency command system is 0.5. Based on the aforementioned channel preferences, obtain the list of entities for receiving early warning notifications. The notification channels preferred by each subject in the early warning notification are associated with corresponding attributes, and a subject preference channel-attribute mapping library is established.

[0012] Preferably, each notification channel in the notification channel set is defined. The channel adaptability is Among them, channel adaptability weight , List of entities issuing early warnings The main body of the early warning notification Notification channels in the notification channel set preference ; Based on the subject preference channel-attribute mapping library, for the list of subjects for the early warning notification Each warning notification subject in Calculate each notification channel in the notification channel set C. Channel compatibility The channel adaptability of all notification channels is sorted from largest to smallest, and the notification channels corresponding to the top M channel adaptability are selected and summarized to generate the target push channel set. Summary of the list of entities issuing early warnings The target push channel set corresponding to each early warning notification subject is used to generate the target push channel combination.

[0013] A power grid monitoring risk early warning push system, the system comprising: The entity information extraction module is used to obtain the risk warning description text generated by the power grid monitoring system. It uses a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules to extract the equipment entity information in the risk warning description text. The entity list output module is used to combine the entity attribute data in the entity information database, sequentially perform basic rule filtering, machine learning optimization matching and conflict verification and correction on the device entity information, and output a warning notification entity list; The channel combination determination module is used to determine the target push channel combination corresponding to the list of warning notification subjects based on the list of warning notification subjects, combined with the channel attribute category and subject channel preference, through channel adaptability calculation, channel status verification, and push execution and feedback. The early warning information push module is used to sort the early warning notification subject list by push priority to generate an early warning notification subject push list, and in combination with the target push channel combination, push risk early warning information to each early warning notification subject in the early warning notification subject push list in sequence, track the reception status of each early warning notification subject in real time, and perform timeout re-push and upgrade handling.

[0014] Compared with related technologies, the power grid monitoring risk early warning push method and system provided by the present invention have the following beneficial effects: This invention acquires risk warning description text generated by a power grid monitoring system, and uses a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules to extract equipment entity information from the risk warning description text. Combining this with entity attribute data from a entity information database, the invention sequentially performs basic rule filtering, machine learning optimization matching, and conflict verification and correction on the equipment entity information, outputting a list of warning notification entities. Based on this list, and considering channel attribute categories and entity channel preferences, the invention determines the target push channel combination corresponding to the warning notification entity list through channel adaptability calculation, channel status verification, and push execution and feedback. The invention then prioritizes the push notifications to generate a push list of warning notification entities, and, based on the target push channel combination, sequentially pushes risk warning information to each entity in the push list. It tracks the reception status of each entity in real time and performs timeout re-push and escalation handling, thereby improving the accuracy and real-time nature of warning push notifications and optimizing the efficiency of power grid risk management.

[0015] This invention significantly improves the accuracy and real-time performance of early warning push notifications through network neural entity recognition, precise subject matching, adaptive channel selection, and time-series priority control. In this invention, entity extraction combined with domain fine-tuning and regularization rules effectively adapts to novel early warning descriptions and reduces the mismatch rate in small-sample scenarios. Subject matching, through hard rule filtering and machine learning optimization, ensures accurate correspondence of responsible subjects. Multi-channel adaptation and real-time verification mechanisms guarantee reliable emergency early warning delivery, automatic switching in case of channel failure, and significantly improved coverage. Time-series control and blockchain evidence storage enable traceability of responsibility, and the timeout escalation mechanism further shortens the response cycle. This invention can adapt to complex scenarios across departments and regions, has strong versatility, and effectively solves the problems of inaccurate matching, low efficiency, unreliable delivery, and ambiguous responsibility in traditional power grid monitoring risk early warning push methods, providing efficient technical support for power grid risk management. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a power grid monitoring risk early warning push method provided in an embodiment of the present invention; Figure 2 A system block diagram of a power grid monitoring risk early warning push system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 The diagram shown is a flowchart of a power grid monitoring risk early warning push method provided by an embodiment of the present invention. Figure 1The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S4 are detailed as follows: S1, Obtain the risk warning description text generated by the power grid monitoring system, and use a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules to extract the equipment entity information in the risk warning description text; S2, combining the subject attribute data in the subject information database, perform basic rule filtering, machine learning optimization matching and conflict verification correction on the device entity information in sequence, and output a list of warning notification subjects; S3. Based on the list of entities receiving the early warning notification, and in combination with the channel attribute category and the entity's channel preference, the target push channel combination corresponding to the list of entities receiving the early warning notification is determined through channel adaptability calculation, channel status verification, and push execution and feedback. S4, sort the list of entities receiving early warning notifications by push priority to generate a push list of entities receiving early warning notifications, and combine it with the target push channel combination to push risk warning information to each entity receiving early warning notifications in the push list in sequence, track the reception status of each entity receiving early warning notifications in real time and perform timeout re-push and upgrade handling.

[0019] In practical applications, risk warning descriptions generated by power grid monitoring systems are collected. A technical solution integrating a pre-trained neural network model with a domain keyword dictionary and regular expression matching rules is employed to accurately extract equipment entity information from these descriptions. By adapting and optimizing the pre-trained neural network model for the power grid domain, its ability to understand text in specialized scenarios is improved. Combined with specifically designed regular expression matching rules and a keyword dictionary, it can efficiently identify and extract structured entity information such as equipment type, equipment name, voltage level, equipment number, and region. This effectively solves the problem of inconsistent textual expressions such as abbreviations and colloquial descriptions, ensuring the completeness and accuracy of equipment entity information.

[0020] Furthermore, based on the extracted device entity information and the associated entity attribute data in the entity information database, the process sequentially performs basic rule-based matching and filtering, machine learning model optimization and matching, and conflict verification and correction, ultimately generating and outputting a list of entities for early warning notifications. The basic rule layer performs initial filtering based on core conditions such as the entity's jurisdiction, responsibility boundaries, and early warning risk level; the machine learning model achieves accurate matching between entities and early warnings through multi-dimensional feature analysis; the conflict verification and correction stage eliminates duplicate entities, handles anomalies such as empty lists, and ensures that at least the core responsible entities are matched through rule downgrading mechanisms, guaranteeing the accuracy and full coverage of notification entities.

[0021] Next, based on the list of entities issuing early warning notifications, and combining the notification channel attribute parameters with the entity's channel preference characteristics, the system determines the target push channel combination for each entity through quantified calculation of channel adaptation, real-time channel status verification, and push execution and status feedback mechanisms. Channel attribute parameters cover key indicators such as delivery timeliness, reachability, and interference level, while entity channel preferences are formed based on historical push data mining. The system comprehensively calculates channel adaptability to select the optimal channel, verifies the online status of channels in real time and dynamically switches faulty channels, and employs push retry and alternative channel switching mechanisms to ensure reliable delivery of early warning information and achieve optimized allocation of channel resources.

[0022] Finally, the list of entities receiving early warning notifications is prioritized and quantified, taking into account factors such as the weight of the entity's responsibilities and matching degree to generate an orderly list of entities to receive early warning notifications. Combined with the target push channel combination, risk warning information is accurately pushed to each entity according to the order of the early warning notification entity list. During the push process, the information reception status of each entity is tracked in real time, and a timeout judgment standard based on the warning level is established. For entities that do not respond, a timeout re-push mechanism and a tiered escalation handling process are initiated. First, the reach is strengthened by escalating the push channel combination; if there is still no response, the notification is pushed to the entity's superior and copied to relevant regulatory departments, ensuring a timely response and closed-loop handling of early warning information and improving the efficiency of power grid risk warning handling.

[0023] In the specific implementation process, the method of using a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules to extract device entity information from the risk warning description text specifically includes: The risk warning description text is subjected to noise reduction, word segmentation and standardization processing to unify the abbreviation and colloquial expression format of the device name in the risk warning description text; Obtain equipment entity annotation data in the power grid field and input it into the network neural pre-training model. Perform domain adaptation fine-tuning on the network neural pre-training model, and process the risk warning description text based on the network neural pre-training model to output entity recognition results. A preset keyword dictionary and regular expression matching rules are used to supplement and verify the entity recognition results, and the regular expression matching results are output. The entity recognition result and the regular expression matching result are fused to form structured device entity information, wherein the device entity information includes device type, device name, device voltage level, device number, device region, and device importance.

[0024] Understandably, the first step is to perform text preprocessing, which involves denoising and removing redundant information, word segmentation to achieve semantic segmentation, and standardizing and unifying the data format. The focus is on standardizing the abbreviation of the device name and colloquial expressions to eliminate recognition interference caused by differences in text expression.

[0025] Subsequently, domain adaptation and entity recognition were performed. Entity annotation data for power grid-specific equipment was obtained and input into the pre-trained neural network model for domain adaptation and fine-tuning, enhancing the model's understanding of power grid terminology and equipment descriptions. The optimized model was then used to perform semantic parsing on the pre-processed warning text, outputting preliminary entity recognition results.

[0026] Then, based on the pre-built dictionary of keywords for power grid equipment and regular expression matching rules, including the rule that “kV” followed by text or letters is used as the line name for voltage level, and the rule that “#” followed by numbers is used as the main transformer number for main transformer equipment, the entity recognition results output by the model are verified and corrected for the second time to generate regular expression matching results.

[0027] Finally, the model recognition results and regular expression matching results are cross-validated and information is fused to remove conflicting data and supplement missing information, forming structured equipment entity information that includes equipment type, equipment name, equipment voltage level, equipment number, equipment region, and equipment importance.

[0028] Extract the equipment type T, equipment voltage level V, equipment region R, and equipment importance S from the equipment entity information, and obtain the risk level L of the warning-marked equipment. The risk level L of the warning-marked equipment is divided into 4 levels, and the higher the risk level L of the warning-marked equipment, the greater the risk of the warning-marked equipment. Retrieve the subject attribute data of all subjects from the subject information database, wherein the subject attribute data includes subject department type D and subject jurisdiction area. Scope of Main Responsibilities The frequency of historical collaboration among entities is Y, and all entities in the entity information database are set as follows: m is the total number of entities in the entity information database; For the aforementioned main jurisdiction area The scope of responsibilities of the main body The encoding conversion is performed, and the historical coordination frequency Y is normalized and preprocessed. Establish the association dimension between the device entity information and the main attribute data to form a dataset to be screened.

[0029] First, from the acquired structured equipment entity information, the equipment type, voltage level, region, and importance are accurately extracted, and the equipment risk level marked by the early warning system is obtained simultaneously. This risk level is divided into four levels, with higher levels indicating a more severe risk to the equipment.

[0030] Subsequently, attribute data of all entities are comprehensively retrieved from the entity information database, covering key information such as entity department type, entity jurisdiction, entity scope of responsibility, and entity historical collaboration frequency, to clarify the total number of entities in the entity information database and define the data range for matching operations.

[0031] Next, the non-numerical attributes of the subject's jurisdiction and scope of responsibility are encoded and converted into standardized data that can be used for algorithm calculations. At the same time, the subject's historical collaboration frequency is normalized and preprocessed to eliminate the computational interference caused by differences in data units, ensuring that all data are comparable and adaptable.

[0032] Finally, based on the core attributes of the device entity information and the key dimensions of the main attribute data, a multi-dimensional association mapping relationship is established, and integrated to form a unified and dimensionally aligned dataset to be screened, providing standardized data input for subsequent basic rule screening.

[0033] Perform basic rule filtering on the dataset to be filtered, specifically including: A pre-defined hard rule base for subject matching is provided, and the hard rule base includes 3 hard rules; Hard rule 1 is applied to the dataset to be filtered to determine the jurisdiction of each subject. Whether the region R to which the device belongs is included, i.e., whether the condition is met. Filter subjects that meet the hard rule 1; For entities that meet the requirements of hard rule 1, hard rule 2 is executed to determine the scope of responsibility of each entity. Whether the device type T is included, i.e., whether the condition is met. Filter subjects that meet the hard rule 2 and add them to the candidate subject set. ; Hard rule 3 is that when the risk level L of the warning labeling device is ≥3, the superior competent authority shall be directly added to the candidate entity set. .

[0034] Understandably, a pre-defined hard rule base for subject matching is used. This rule base contains three core hard rules, which provide a clear logical basis for the initial screening of subjects and ensure the standardization and accuracy of the screening process.

[0035] When executing hard rule 1, for each subject in the dataset to be filtered, it is determined whether its jurisdiction fully includes the area to which the early warning device belongs. Through spatial range adaptability verification, subjects with regional jurisdiction are filtered out, and subjects without relevant jurisdiction are removed.

[0036] For entities that meet Hard Rule 1, Hard Rule 2 is applied to determine whether the scope of responsibility of each entity covers the type of current early warning equipment. Through the responsibility boundary adaptability check, entities with equipment management responsibilities are further screened out and included in the candidate entity set.

[0037] When implementing hard rule 3, if the risk level of the early warning device reaches level 3 or above, the high-risk level subject expansion mechanism will be activated, directly including the corresponding superior competent department into the candidate subject set, to ensure that high-risk early warnings can receive hierarchical attention and handling, forming the final candidate subject set.

[0038] For the candidate subject set Performing machine learning to optimize matching specifically includes: Five feature variables were defined, including the jurisdictional matching degree between the subject and the device. The correlation between the main responsibilities and the types of early warnings Risk level normalized value Normalized value of historical collaboration frequency Response time normalized value ; Among them, the jurisdiction matching degree between the subject and the equipment If the candidate subject set If the candidate subject and the device are a perfect match, then If the candidate subject partially matches the device, then If the candidate entity and the device are indirectly matched, then ; Regarding the correlation between the main responsibilities and the warning types If the primary responsibility of the candidate entity is directly related to the type of early warning, then If the primary responsibility of the candidate entity and the type of early warning are related, then If the primary responsibility of the candidate entity and the type of early warning are related as auxiliary responsibilities, then ; For the risk level normalized value If the risk level L of the warning labeling device is level 1, then If the risk level L of the warning labeling device is level 2, then If the risk level L of the warning labeling device is level 3, then If the risk level L of the warning labeling device is level 4, then ; For any candidate subject The normalized value of the historical collaboration frequency That is, the candidate subject Number of similar warnings processed in history The normalized value, corresponding to the calculation formula is: n is the set of candidate subjects The total number of candidate entities; For the normalized value of response time If the historical average response time of the candidate entity is less than 5 minutes, then If the historical average response time of the candidate entity is 5 to 15 minutes, then If the historical average response time of the candidate entity is greater than 15 minutes, then ; The matching degree between the candidate subjects and this early warning is calculated using a logistic regression algorithm. Among them, the matching degree weight ; In the candidate subject set In this process, candidate entities with a matching score greater than or equal to 0.6 are first added to the list of entities to be verified. Then, the matching scores of other candidate entities are sorted from largest to smallest. The candidate entities corresponding to the top N matching scores are selected and added to the list of entities to be verified. The value of N is dynamically adjusted according to the risk level L of the warning labeling device.

[0039] Five feature variables are set as the core dimensions for matching degree calculation: the matching degree of the subject and the equipment, the correlation between the subject's responsibilities and the warning type, the normalized value of the risk level, the normalized value of the historical collaboration frequency, and the normalized value of the response time, which comprehensively cover multiple dimensions such as spatial adaptation, responsibility adaptation, risk adaptation, experience adaptation, and efficiency adaptation.

[0040] For the matching degree of jurisdiction between the subject and the equipment, it is divided into three levels according to the degree of adaptation of the jurisdiction of the subject and the equipment: the highest adaptation score is taken when there is a complete match, the medium adaptation score is taken when there is a partial match, and the basic adaptation score is taken when there is an indirect match. For the relevance between the subject's responsibilities and the warning type, it is divided into three levels according to the degree of responsibility association: direct responsibility, related responsibility, and auxiliary responsibility, each corresponding to a different relevance score. For the risk level normalization value, a gradient normalization score is assigned according to the four-level risk level of the warning equipment, with higher risk levels resulting in higher scores. For the historical collaboration frequency normalization value, the normalization result of the historical collaboration frequency of each candidate subject is calculated based on the highest number of times that all subjects in the candidate subject set have historically handled the same type of warning. For the response time normalization value, it is divided into three intervals according to the historical average response time of the candidate subjects, with shorter response times corresponding to higher normalization scores.

[0041] The matching degree calculation uses the logistic regression algorithm, which integrates the values ​​of five feature variables through a weighted summation formula to ensure that the core adaptation dimension occupies a higher weight.

[0042] The candidate subject selection follows a dual standard. First, candidate subjects with a matching score of 0.6 or above are directly included in the list of subjects to be verified. Then, the remaining candidate subjects are sorted from high to low according to their matching scores. The number of selected subjects is dynamically adjusted according to the risk level of the early warning equipment. The top N candidate subjects are added to the list of subjects to be verified to ensure that the list contains both highly compatible subjects and comprehensive matching.

[0043] Perform conflict verification and correction on the list of entities to be verified, and output the list of entities to issue early warning notifications, specifically including: Perform hierarchical verification on the entities to be verified in the list of entities to be verified. If entities to be verified at different levels in the same department are matched repeatedly, only the entities at the highest level are retained, and a list of deduplicated entities is output. Determine if the deduplication subject list is empty. If it is, perform rule downgrading on the hard rule base and re-execute basic rule matching and machine learning optimization matching until the deduplication subject list is not empty, then output the warning notification subject list. .

[0044] It should be noted that, firstly, all entities in the list to be verified are verified by administrative level. If there are duplicate matches between entities at different levels in the same department, only the entity at the highest level is retained to avoid confusion caused by redundancy in levels, and a deduplicated entity list is output.

[0045] Next, it is determined whether the deduplication subject list is empty. If the deduplication subject list is empty, the rule degradation mechanism is activated, and the restrictions such as jurisdiction and scope of responsibility in the basic rules are appropriately relaxed. The basic rule screening and machine learning optimization matching process is re-executed until a non-empty deduplication subject list is generated, ensuring that at least one core responsible subject is matched.

[0046] Finally, the deduplicated subject list after deduplication and rule downgrading is finally verified to ensure that the subject information is not duplicated, missing, or logically conflicting. The final list of warning notification subjects is then output to provide accurate subject object basis for subsequent warning push channel selection.

[0047] Build a collection of notification channels ,in, Indicates the scheduling screen, Indicates mobile app, Indicates text message, Indicates telephone, Indicates the emergency command system; Combined with the aforementioned channel attribute categories, including delivery timeliness Accessibility Interference Define the attributes of each notification channel in the notification channel set, wherein, for the delivery timeliness... The delivery time for the dispatch screen is 5 seconds, the delivery time for the mobile APP is 3 seconds, the delivery time for the SMS is 10 seconds, the delivery time for the phone call is 1 second, and the delivery time for the emergency command system is 8 seconds; regarding the reachability... The reachability of the dispatch screen is 90%, the reachability of the mobile APP is 95%, the reachability of SMS is 98%, the reachability of telephone is 99%, and the reachability of the emergency command system is 85%; regarding the interference level... The interference level of the dispatch screen is 0.3, the interference level of the mobile APP is 0.6, the interference level of the SMS is 0.7, the interference level of the telephone is 0.9, and the interference level of the emergency command system is 0.5. Based on the aforementioned channel preferences, obtain the list of entities for receiving early warning notifications. The notification channels preferred by each subject in the early warning notification are associated with corresponding attributes, and a subject preference channel-attribute mapping library is established.

[0048] Define each notification channel in the set of notification channels. The channel adaptability is Among them, channel adaptability weight , List of entities issuing early warnings The main body of the early warning notification Notification channels in the notification channel set preference ; Based on the subject preference channel-attribute mapping library, for the list of subjects for the early warning notification Each warning notification subject in Calculate each notification channel in the notification channel set C. Channel compatibility The channel adaptability of all notification channels is sorted from largest to smallest, and the notification channels corresponding to the top M channel adaptability are selected and summarized to generate the target push channel set. Summary of the list of entities issuing early warnings The target push channel set corresponding to each early warning notification subject is used to generate the target push channel combination.

[0049] First, a multi-channel notification system covering power grid early warning push scenarios is constructed, including dispatch dashboards, mobile apps, SMS, telephone, and emergency command systems. This system comprehensively covers push needs in different scenarios, ensuring efficient delivery of routine early warnings as well as immediate access to emergency warnings, laying the foundation for accurate adaptation in the future.

[0050] For each notification channel within the notification channel set, quantitative parameters are defined for delivery timeliness, reachability, and interference. Regarding delivery timeliness, the shortest is 1 second for phone calls, 3 seconds for mobile applications, 5 seconds for dispatch dashboards, 8 seconds for emergency command systems, and the longest is 10 seconds for SMS. For reachability, phone calls reach a maximum of 99%, SMS 98%, mobile applications 95%, dispatch dashboards 90%, and emergency command systems 85%. Regarding interference, the values ​​are 0.9 for phone calls, 0.7 for SMS, 0.6 for mobile applications, 0.5 for emergency command systems, and 0.3 for dispatch dashboards; higher interference indicates a greater likelihood of attracting the recipient's attention. These precise quantitative attributes provide data support for channel compatibility assessment.

[0051] Based on historical push data mining, the channel usage preferences of each warning notification subject in the list of warning notification subjects are obtained, clarifying the commonly used and preferred notification channels for different warning notification subjects. Subject preferences are then correlated and matched with the core attribute parameters of the corresponding channels to establish a structured subject preference channel-attribute mapping library. This enables precise matching between subject needs and channel characteristics, providing a data foundation for personalized channel adaptation.

[0052] Furthermore, a channel adaptability assessment is constructed, comprehensively considering four dimensions: delivery timeliness, reachability, interference level, and subject preference. Delivery timeliness accounts for 40% of the weight, reachability for 30%, interference level for 20%, and subject preference for 10%. Subject preference for channels is categorized into three levels based on usage frequency: frequently used channels have the highest preference, followed by occasionally used channels, and unused channels have the lowest preference, comprehensively covering both objective attributes and subjective needs.

[0053] Based on the subject preference channel-attribute mapping library, for each warning notification subject in the list of warning notification subjects, the channel adaptability calculation formula is substituted into the database to calculate the adaptability of all notification channels in the notification channel set. The notification channels are sorted from high to low adaptability. According to the warning level and actual needs, the top M notification channels with the best adaptability are selected to form the target push channel set for that warning notification subject, ensuring that the channels are highly consistent with the subject's needs.

[0054] Finally, the target push channel sets corresponding to all warning notification subjects in the warning notification subject list are aggregated, redundant and duplicate channels are eliminated, and a target push channel combination covering all warning notification subjects is formed. This combination not only ensures the personalized push needs of individual warning notification subjects, but also optimizes the allocation of channel resources, ensuring that different types of warning information accurately reach the corresponding warning notification subjects through the optimal channel combination, providing a clear channel plan for subsequent push execution.

[0055] like Figure 2 The diagram shown is a system block diagram of a power grid monitoring risk early warning push system provided by an embodiment of the present invention. The system includes: The entity information extraction module is used to obtain the risk warning description text generated by the power grid monitoring system. It uses a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules to extract the equipment entity information in the risk warning description text. The entity list output module is used to combine the entity attribute data in the entity information database, sequentially perform basic rule filtering, machine learning optimization matching and conflict verification and correction on the device entity information, and output a warning notification entity list; The channel combination determination module is used to determine the target push channel combination corresponding to the list of warning notification subjects based on the list of warning notification subjects, combined with the channel attribute category and subject channel preference, through channel adaptability calculation, channel status verification, and push execution and feedback. The early warning information push module is used to sort the early warning notification subject list by push priority to generate an early warning notification subject push list, and in combination with the target push channel combination, push risk early warning information to each early warning notification subject in the early warning notification subject push list in sequence, track the reception status of each early warning notification subject in real time, and perform timeout re-push and upgrade handling.

[0056] Figure 2 The apparatus of the illustrated embodiment can be used to perform corresponding actions. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0057] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the steps of the power grid monitoring risk early warning push method as described in any of the above.

[0058] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device according to an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32, and a computer program; wherein... The memory 32 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0059] Processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0060] Alternatively, the memory 32 can be either standalone or integrated with the processor 31.

[0061] When the memory 32 is a device independent of the processor 31, the device may further include: Bus 33 is used to connect the memory 32 and the processor 31.

[0062] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of a power grid monitoring risk early warning push method as described in any of the above claims.

[0063] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0064] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0065] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0066] Through the above embodiments, this invention, through a power grid monitoring risk warning push method and system, obtains risk warning description text generated by the power grid monitoring system, and uses a network neural pre-trained model combined with a keyword dictionary and regular expression matching rules to extract equipment entity information from the risk warning description text; combined with the subject attribute data in the subject information database, it sequentially performs basic rule filtering, machine learning optimization matching, and conflict verification and correction on the equipment entity information, outputting a list of warning notification subjects; based on the list of warning notification subjects, combined with channel attribute categories and subject channel preferences, it determines the target push channel combination corresponding to the list of warning notification subjects through channel adaptability calculation, channel status verification, and push execution and feedback; it performs push priority sorting on the list of warning notification subjects to generate a push list of warning notification subjects, and sequentially pushes risk warning information to each warning notification subject in the push list in combination with the target push channel combination, tracks the reception status of each warning notification subject in real time, and performs timeout re-push and upgrade handling, thereby improving the accuracy of warning push subjects and the real-time performance of notifications, and optimizing the efficiency of power grid risk handling.

[0067] This invention significantly improves the accuracy and real-time performance of early warning push notifications through network neural entity recognition, precise subject matching, adaptive channel selection, and time-series priority control. In this invention, entity extraction combined with domain fine-tuning and regularization rules effectively adapts to novel early warning descriptions and reduces the mismatch rate in small-sample scenarios. Subject matching, through hard rule filtering and machine learning optimization, ensures accurate correspondence of responsible subjects. Multi-channel adaptation and real-time verification mechanisms guarantee reliable emergency early warning delivery, automatic switching in case of channel failure, and significantly improved coverage. Time-series control and blockchain evidence storage enable traceability of responsibility, and the timeout escalation mechanism further shortens the response cycle. This invention can adapt to complex scenarios across departments and regions, has strong versatility, and effectively solves the problems of inaccurate matching, low efficiency, unreliable delivery, and ambiguous responsibility in traditional power grid monitoring risk early warning push methods, providing efficient technical support for power grid risk management.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power grid monitoring risk early warning pushing method, characterized in that, The method comprises: obtaining the risk warning description text generated by the power grid monitoring system, using a network neural pre-training model combined with a keyword dictionary and a regular matching rule to extract the equipment entity information in the risk warning description text; combining the subject attribute data in the subject information library, sequentially performing basic rule screening, machine learning optimized matching and conflict verification correction on the equipment entity information, and outputting a warning notification subject list; based on the warning notification subject list, combining the channel attribute category and the subject channel preference, determining the target push channel combination corresponding to the warning notification subject list through channel adaptation degree calculation, channel state verification and push execution and feedback; performing push priority sorting on the warning notification subject list to generate a warning notification subject push list, and combining the target push channel combination, sequentially pushing risk warning information to each warning notification subject in the warning notification subject push list, and tracking the receiving state of each warning notification subject in real time and performing timeout re-push and escalation disposal.

2. The power grid monitoring risk early warning pushing method according to claim 1, characterized in that, The network neural pre-training model combined with the keyword dictionary and the regular matching rule is used to extract the equipment entity information in the risk warning description text, specifically comprising: performing denoising, word segmentation and standardization processing on the risk warning description text, and unifying the abbreviation and colloquial expression format of the equipment name in the risk warning description text; obtaining the equipment entity annotation data in the power grid field and inputting it into the network neural pre-training model, performing field adaptation fine-tuning on the network neural pre-training model, and processing the risk warning description text based on the network neural pre-training model to output entity recognition results; presetting a keyword dictionary and a regular matching rule, and performing supplementary verification on the entity recognition results to output regular matching results; fuse the entity recognition results and the regular matching results to form structured equipment entity information, wherein the equipment entity information includes equipment type, equipment name, equipment voltage level, equipment number, equipment belonging area, and equipment importance.

3. The power grid monitoring risk early warning pushing method according to claim 2, characterized in that, extract the equipment type T, equipment voltage level V, equipment belonging area R and equipment importance S in the equipment entity information, and obtain the warning labeled equipment risk level L, wherein the warning labeled equipment risk level L is divided into 4 levels, and the higher the warning labeled equipment risk level L, the greater the warning labeled equipment risk; retrieve the subject attribute data of all subjects from the subject information base, wherein the subject attribute data comprises subject department type D, subject jurisdiction area , subject responsibility range , subject history coordination frequency Y, and set all subjects in the subject information base as , and m is the total number of subjects in the subject information base; The main body jurisdiction The main body responsibility range Encoding conversion is performed, and the historical coordination frequency Y is normalized and pretreated. establish the association dimension of the equipment entity information and the subject attribute data to form a to-be-screened data set.

4. The power grid monitoring risk early warning pushing method according to claim 3, characterized in that, performing basic rule screening on the to-be-screened data set, specifically comprising: presetting a hard rule library for subject matching, and the hard rule library includes 3 hard rules; Apply hard rule 1 to the dataset to be filtered to determine the jurisdiction of each subject. Whether the region R to which the device belongs is included, i.e., whether the condition is met. Filter subjects that meet the hard rule 1; performing hard rule 2 on the subjects meeting the hard rule 1 to determine the subject responsibility scope of each subject whether the device type T is included, i.e. whether the condition is met, screening the subjects meeting the hard rule 2 and adding them to the candidate subject set ; Hard rule 3 is that when the early warning label device risk level L≥3, the superior supervisory department is directly added to the candidate subject set .

5. The power grid monitoring risk early warning pushing method according to claim 4, characterized in that, to the candidate subject set performing machine learning optimization matching, specifically comprising: Set 5-dimensional characteristic variables, including the jurisdiction matching degree between the subject and the device , the correlation between the subject responsibility and the early warning type , the risk level normalized value , the historical coordination frequency normalized value , the response timeliness normalized value ; wherein, for the subject and device jurisdiction matching degree , if the candidate subject in the candidate subject set is completely matched with the device, then ; if the candidate subject is partially matched with the device, then ; if the candidate subject is indirectly matched with the device, then ; For the relevance of the subject responsibility and the early warning type , if the subject responsibility of the candidate subject and the early warning type are in a direct responsibility relationship, then ; if the subject responsibility of the candidate subject and the early warning type are in an associated responsibility relationship, then ; if the subject responsibility of the candidate subject and the early warning type are in an auxiliary responsibility relationship, then ; for the risk level normalized value if the pre-warning marking device risk level L is level 1, then if the pre-warning marking device risk level L is level 2, then if the pre-warning marking device risk level L is level 3, then if the pre-warning marking device risk level L is level 4, then if the pre-warning marking device risk level L is level 5, then For any candidate subject The normalized value of the historical collaboration frequency That is, the candidate subject Number of similar warnings processed in history The normalized value, corresponding to the calculation formula is: n is the set of candidate subjects The total number of candidate entities; for the response age normalized value if the candidate subject historical average response time is less than 5 minutes, then if the candidate subject historical average response time is between 5 and 15 minutes, then if the candidate subject historical average response time is greater than 15 minutes, then ; A logic regression algorithm is used to calculate the matching degree of the candidate subject and the current early warning wherein the matching degree weight ; In the candidate subject set In the candidate subject set, the candidate subjects with the matching degree Score greater than or equal to 0.6 are added to a to-be-verified subject list, and the matching degree Scores of other candidate subjects are sorted from large to small, and the first N candidate subjects corresponding to the matching degree Scores are selected and added to the to-be-verified subject list, wherein the value of N is dynamically adjusted according to the risk level L of the early warning labeling device.

6. The power grid monitoring risk early warning pushing method according to claim 5, characterized in that, performing conflict verification correction on the to-be-verified subject list to output the warning notification subject list, specifically comprising: performing hierarchical verification on the to-be-verified subjects in the to-be-verified subject list, if the same department and different levels of to-be-verified subjects are repeatedly matched, only the to-be-verified subject of the highest level is retained, and a de-duplication subject list is outputted; determining whether the deduplication subject list is empty, if the deduplication subject list is empty, performing rule degradation processing on the hard rule library, and re-executing basic rule matching and machine learning optimization matching until the deduplication subject list is not empty, and outputting the early warning notification subject list .

7. The power grid monitoring risk early warning pushing method according to claim 6, characterized in that, Constructing a notification channel set Wherein, Indicate dispatch large screen, Indicate mobile APP, Indicate SMS, Indicate phone, Indicate emergency command system; in combination with the channel attribute category, including delivery time , reachability , interference degree , define the attributes of each notification channel in the notification channel set, wherein for the delivery time , the delivery time of the dispatch large screen is 5 seconds, the delivery time of the mobile terminal APP is 3 seconds, the delivery time of the short message is 10 seconds, the delivery time of the telephone is 1 second, and the delivery time of the emergency command system is 8 seconds; for the reachability , the reachability of the dispatch large screen is 90%, the reachability of the mobile terminal APP is 95%, the reachability of the short message is 98%, the reachability of the telephone is 99%, and the reachability of the emergency command system is 85%; for the interference degree , the interference degree of the dispatch large screen is 0.3, the interference degree of the mobile terminal APP is 0.6, the interference degree of the short message is 0.7, the interference degree of the telephone is 0.9, and the interference degree of the emergency command system is 0.5; Based on the subject channel preference, the early warning notification subject list is acquired The notification channel of each early warning notification subject preference is associated with the corresponding attribute, and a subject preference channel-attribute mapping library is established.

8. The power grid monitoring risk early warning pushing method according to claim 7, characterized in that, define each notification channel in the set of notification channels a channel fitness , wherein the channel fitness weight , represents a list of early warning notification subjects in the set of early warning notification subjects a preference for a notification channel in the set of notification channels , ; Based on the subject preference channel-attribute mapping library, for the list of subjects for the early warning notification Each warning notification subject in Calculate each notification channel in the notification channel set C. Channel compatibility The channel adaptability of all notification channels is sorted from largest to smallest, and the notification channels corresponding to the top M channel adaptability are selected and summarized to generate the target push channel set. Summary of the list of entities issuing early warnings The target push channel set corresponding to each early warning notification subject is used to generate the target push channel combination.

9. A power grid monitoring risk early warning pushing system applied to the power grid monitoring risk early warning pushing method of any one of claims 1-8, characterized in that, The system comprises: The entity information extraction module is configured to acquire risk early warning description text generated by the power grid monitoring system, and extract device entity information in the risk early warning description text by using a network neural pre-training model in combination with a keyword dictionary and a regular matching rule. The subject list output module is configured to execute, in combination with subject attribute data in a subject information library, basic rule screening, machine learning optimized matching, and conflict verification and correction on the device entity information in sequence, and output an early warning notification subject list. The channel combination determination module is configured to determine a target push channel combination corresponding to the early warning notification subject list based on the early warning notification subject list in combination with channel attribute categories and subject channel preferences, through channel adaptation degree calculation, channel state verification, and push execution and feedback. The early warning information push module is configured to execute push priority sorting on the early warning notification subject list to generate an early warning notification subject push list, and in combination with the target push channel combination, push risk early warning information to each early warning notification subject in the early warning notification subject push list in sequence, and track the receiving state of each early warning notification subject in real time and execute timeout re-pushing and escalation disposal.

Images