Switching operation safety checking system and method based on artificial intelligence

Abstract

The invention discloses a switching operation safety checking system based on artificial intelligence, and the system comprises a model building module which obtains an enhanced domain language model through a self-supervised pre-training technology and a power grid dispatching professional feature dictionary, and then obtains a natural language analysis model through the combination with a bidirectional long-short-term memory network; the text analysis module utilizes a natural language analysis model to convert the preprocessed scheduling service natural language text into structured objectified data which identifies operation equipment, an operation equipment state, an operation event, operation rule information, an operation event intention and operation logic information; the safety checking module is used for carrying out topology analysis, image-text comparison, intelligent ticket forming, maintenance linkage, sand table rehearsal and checking before order issuing before switching operation is executed based on objectified data; key link checking, operation equipment state checking and operation event time sequence checking are carried out in the switching operation execution process. According to the invention, the safety and the working efficiency of switching operation are improved.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to an artificial intelligence-based safety verification system and method for switching operations. Background Technology

[0002] With the construction of new power systems, the workload of planned maintenance, temporary maintenance, and commissioning of new equipment has increased year by year, leading to a corresponding increase in the volume of dispatching and switching operations. Currently, most of the operating systems used by the State Grid have functions such as drafting, reviewing, pre-issuing, executing, and archiving operating instructions, but they lack functions for security verification of each stage of the operating instruction ticket, intelligent control of the entire process, and multi-dimensional post-evaluation.

[0003] Specifically, the drafting and reviewing stages lack auxiliary functions, relying heavily on dispatchers' manual experience and lacking intelligent verification of the accuracy of operation instructions and sandbox simulation of switching operations. The execution stage lacks intelligent control, requiring dispatchers to repeatedly verify with the site, resulting in low efficiency of switching operations. After the operation instructions are archived, the drafting, reviewing, pre-issuance, and execution stages are not effectively evaluated, failing to improve the quality of operation instructions and the efficiency of switching operations for dispatchers at all levels in the power grid. Summary of the Invention

[0004] The purpose of this invention is to provide a safety verification system and method for switching operations based on artificial intelligence. This invention improves the safety and efficiency of switching operations.

[0005] To achieve this objective, the present invention provides an artificial intelligence-based safety verification system for switching operations, comprising: The model building module is used to perform self-supervised pre-training on the knowledge corpus of power grid dispatching through masked language model and next sentence prediction task to generate basic language model. The basic language model is then subjected to domain-adaptive training through power grid dispatching professional feature dictionary to obtain enhanced domain language model. The enhanced domain language model is then combined with bidirectional long short-term memory network to obtain natural language parsing model. The text parsing module is used to perform word embedding and position encoding on the preprocessed scheduling business natural language text to obtain an input vector sequence. The input vector sequence is then encoded by a natural language parsing model to obtain an information encoding vector. The information encoding vector is mapped to a sequence label that conforms to the BIO labeling pattern. Based on the sequence label, the module identifies and extracts the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information from the scheduling business natural language text to obtain structured objectified data. The safety verification module is used to perform topology analysis, image and text comparison, intelligent ticketing, maintenance linkage, sand table simulation and pre-order verification based on the objectified data before the switching operation is executed; during the switching operation, it is used to verify key links in the operation instruction ticket, operation equipment status and operation event timing based on the objectified data.

[0006] Preferably, the safety verification module is also used to perform a post-evaluation of the entire process from before the switching operation to the completion of the switching operation based on the objectified data after the switching operation is completed, thereby establishing a full-process control mechanism.

[0007] Preferably, the specific process of constructing the natural language parsing model is as follows: The knowledge corpus in the field of power grid dispatching includes historical maintenance tickets, switching operation instruction tickets, equipment commissioning sheets, dispatching procedures, and equipment models; Based on a general pre-trained model based on the Transformer architecture, a masking language model is used to randomly mask some words in the text of the power grid dispatching domain knowledge corpus. The general pre-trained model based on the Transformer architecture then predicts the masked words, thereby learning the semantics of the words in the text context. The next sentence prediction task allows the general pre-trained model based on the Transformer architecture to determine whether two sentences in the text of the power grid dispatching domain knowledge corpus are in a continuous relationship, thereby generating a basic language model. A dictionary adapter is inserted between the Transformer layers of the basic language model. When the text of the power grid dispatch domain knowledge corpus passes through the basic language model, the dictionary adapter matches the currently processed character vector with the word vector in the power grid dispatch professional feature dictionary, and performs domain-adaptive training through attention mechanism fusion to enhance the feature representation of domain-related words and obtain an enhanced domain language model. The enhanced domain language model is used to obtain the text vector sequence of power grid dispatch domain knowledge corpus. The vector sequence is then input into a bidirectional long short-term memory network to capture long-distance dependencies and sequential relationships in the vector sequence, and finally a natural language parsing model is obtained.

[0008] Preferably, the specific process of performing word embedding and positional encoding on the preprocessed scheduling business natural language text to obtain the input vector sequence is as follows: The scheduling business natural language text is preprocessed. According to the predefined rules that conform to the BIO annotation pattern, each character in the preprocessed scheduling business natural language text is assigned a BIO tag. B represents the first character in the sentence that belongs to an entity word, I represents the character in the sentence that does not belong to any entity word, and O represents the character that does not belong to any entity word. Each character in the preprocessed scheduling business natural language text that has been annotated with BIO tags is converted into a vector through word embedding technology, and a position encoding vector is added to obtain the input vector sequence. If the preprocessed scheduling business natural language text contains sentence L, and sentence L has n characters (i.e., a length of n), then by using word embedding technology to split each character of the sentence and mapping each character to a vector representation, then... ; in Represents the nth word; encoded by positional vector. Indicates the position of corresponding words in a sentence, where, The vector representing the position of the nth character; the input vector sequence is obtained by combining the vector representation of sentence L with the corresponding position encoding vector.

[0009] Preferably, the specific process of encoding the input vector sequence using a natural language parsing model to obtain an information encoding vector, mapping the information encoding vector to a sequence label conforming to the BIO annotation pattern, and identifying and extracting the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information from the natural language text of the scheduling business based on the sequence label is as follows: The input vector sequence is forward computed using the Transformer encoder in a natural language parsing model to obtain the output vector T after Transformer encoding. ; The output vector T is forward computed using a bidirectional long short-term memory (BSSM) network in a natural language parsing model to obtain the information encoding vector H after encoding by the BSSM network. , h n The vector representing the position of the nth character in the sentence; The information encoding vector H is computed forward by the multilayer perceptron classification layer in the natural language parsing model. The probability distribution of BIO labels for each word in the information encoding vector H is calculated by combining the Softmax function. The BIO label with the highest probability is selected as the prediction result for that word, thus mapping the information encoding vector H into a sequence label that conforms to the BIO labeling pattern. Based on the sequence label, the original text at the corresponding position is extracted, thereby identifying and extracting the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information in the natural language text of the scheduling business, and assembling them into structured object data in the form of key-value pairs.

[0010] Preferably, before the switching operation is executed, the specific process of performing topology analysis, image-text comparison, intelligent ticketing, maintenance linkage, sand table simulation, and pre-order verification based on the objectified data is as follows: Topology analysis is performed based on the objectified data. According to the ID of the operating device in the objectified data, it can be queried in the database of the SCADA system. By utilizing the association between topology point numbers, the operating device is automatically located and the topology point number associated with the operating device is obtained. Then, all electrically directly or indirectly related primary devices are found, and the electrical connection relationship between the operating devices is automatically obtained. Based on the objectified data, a graphic comparison is performed. According to the ID of the operating device, the status of the operating device, and the intent of the operating event in the objectified data, the topology analysis function is invoked to generate a current status list of all main devices and their fully associated devices involved in the operation instruction ticket. The status of the operating devices in the list is compared with the actual status of the operating devices on site, and the status of the operating devices is confirmed or anomalies are reported to ensure the accuracy of the status of the operating devices before the ticket is drafted. Based on the objectified data, intelligent ticket generation is performed. After the dispatching and operation personnel draft the operation tasks and method requirements, the operation tasks and method requirements are parsed using a natural language parsing model to extract the operation equipment and operation event intent. The intelligent ticket generation model, which is pre-trained based on a large number of historical typical operation tickets and dispatching procedures, is called to automatically generate a complete operation instruction ticket that conforms to the dispatching specifications and is logically correct, according to the extracted operation equipment and operation event intent. Based on the objectified data, maintenance linkage is carried out. All maintenance tickets to be executed are periodically obtained from the maintenance ticket system. Natural language parsing model is used to determine whether each maintenance ticket needs to be written with corresponding operation instruction tickets and the time window for writing the tickets. When it is detected that the dispatching operator has missed writing the operation instruction ticket within the time window, a pop-up window and a voice alarm based on speech synthesis technology are automatically triggered to remind the operator. Based on the objectified data, a sand table simulation is performed. According to the objectified data of the operation instruction ticket, the state changes of the operating equipment are dynamically simulated, and power flow calculation is automatically triggered to analyze the power grid state after the simulated state changes of the operating equipment. Pre-order verification is performed based on the objectified data. For power outage operations, the verification checks whether the objectified data contains key requirements and verifies whether the associated maintenance ticket has been completed, the associated safety and stability measures have been started, and whether the operation of the superior-approved equipment has been authorized. If any condition is not met, the operation is prohibited. For power restoration operations, the verification checks whether the associated maintenance ticket has been completed, the associated protection setting sheet has been executed, and the associated power restoration commissioning sheet or new equipment commissioning plan has been started. If any condition is not met, the operation is prohibited.

[0011] Preferably, during the switching operation, the specific process of performing key link verification, equipment status verification, and timing verification based on the objectified data is as follows: Based on the objectified data, key steps are verified to identify key operational steps in the operation instruction ticket. When a key operational step is simulated or actually executed, an enhanced verification mechanism is automatically triggered: For charging operations, a secondary confirmation request is sent to the operating units on both sides of the line. The operation instruction is only allowed to be issued after both sides have completed the confirmation; For operations of equipment authorized by the superior, the shift leader authorization mechanism is triggered, and an authorization request interface pops up. The operation instruction can only continue to be executed after the shift leader confirms the authorization. Based on the objectified data, the device status is checked. When an operation command is executed, the actual status of the corresponding operation device is queried in real time according to the operation device and operation device status in the objectified data. If the operation device status in the objectified data is inconsistent with the actual status of the operation device, the command is prohibited and an alarm is triggered. Based on the objectified data, timing verification is performed. When there are multiple operation tasks, timing requirements are parsed from the approval opinions of the associated maintenance tickets. If there are no explicit requirements, power flow calculation simulation is automatically performed to simulate all possible operation sequences and select the optimal timing sequence with the least impact on the power grid.

[0012] Preferably, after the switching operation is completed, a post-evaluation is performed on the entire process from before the switching operation to the completion of the switching operation based on the objectified data, thereby establishing a full-process control mechanism. The specific process is as follows: Based on the objectified data, all objectified data, operation logs, personnel behavior records, timestamps, operating equipment status change sequences and verification results generated in the entire process from before the switching operation to the completion of the switching operation are recorded to form an electronic operation archive. Based on the electronic operation archive and the objectified data, a multi-dimensional post-evaluation analysis is performed to obtain the post-evaluation results. Based on the post-evaluation results, typical error cases, high-risk operations, and weak links are selected for assessment and training of dispatching and operation personnel. New typical operation patterns, optimized operation logic, and common types of misoperations discovered in the post-evaluation results are used as feedback data to incrementally learn and iteratively optimize the knowledge base of the intelligent ticketing model and natural language parsing model. By mining historical operation data, potential systemic risk points and rule loopholes are identified, optimization suggestions are automatically generated, and the rule base update process is triggered, thereby achieving comprehensive intelligent control of switching operations.

[0013] A safety verification method for switching operations based on artificial intelligence includes the following steps: Self-supervised pre-training of knowledge corpus in the power grid dispatching domain is performed using a masked language model and a next sentence prediction task to generate a basic language model. The basic language model is then subjected to domain-adaptive training using a power grid dispatching professional feature dictionary to obtain an enhanced domain language model. Finally, the enhanced domain language model is combined with a bidirectional long short-term memory network to obtain a natural language parsing model. Word embedding and position encoding are performed on the preprocessed scheduling business natural language text to obtain an input vector sequence. The input vector sequence is then encoded using a natural language parsing model to obtain an information encoding vector. The information encoding vector is mapped to a sequence label that conforms to the BIO annotation pattern. Based on the sequence label, the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information in the scheduling business natural language text are identified and extracted to obtain structured objectified data. Before the switching operation is executed, topology analysis, image and text comparison, intelligent ticketing, maintenance linkage, sand table simulation and pre-order verification are performed based on the objectified data; during the switching operation, key links in the operation instruction ticket, operation equipment status verification and operation event timing verification are performed based on the objectified data.

[0014] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0015] The beneficial effects of this invention are: This invention utilizes a natural language processing model to perform multiple safety checks on the entire process of switching operations before, during, and after execution. It automatically identifies and intercepts non-compliant operations, effectively preventing misoperations and omissions, and ensuring the stable operation of the power grid. Through intelligent ticketing, automated verification, and real-time data querying, this invention improves the efficiency of switching operations and reduces the workload of dispatchers. By simulating power flow changes, associating maintenance ticket information, and conducting multi-dimensional post-evaluation, this invention dynamically optimizes the operation logic, reduces the probability of operational accidents, and effectively ensures the safety of dispatching switching operations. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Example 1 An artificial intelligence-based safety verification system for switching operations, such as... Figure 1 As shown, it includes: The model building module is used to perform self-supervised pre-training on a corpus of knowledge related to power grid dispatching using a masked language model (which trains the basic language model to understand the contextual semantics of language by having it predict words that are intentionally masked in a text) and a next-sentence prediction task (which determines whether two sentences are in a continuous context in the original text). This generates a basic language model. The basic language model is then subjected to domain-adaptive training using a power grid dispatching feature dictionary to obtain an enhanced domain language model (LEBERT, i.e., Lexicon Enhanced BERT). Finally, the enhanced domain language model is connected to a Bidirectional Long Short-Term Memory (BiLSTM) network. The LEBERT-BiLSTM model is obtained by combining the domain language model with the power grid dispatch knowledge corpus. This design is self-supervised pre-trained using the power grid dispatch domain knowledge corpus to adapt to the vocabulary, syntax and context of the power field. The accuracy of entity recognition is improved by inserting a power grid dispatch professional feature dictionary. By combining the enhanced domain language model with the bidirectional long short-term memory network, it can not only understand the meaning of each word, but also capture the sequence and long-distance dependencies between operation steps, and accurately parse the intention of operation events and operation logic information. The text parsing module performs word embedding and position encoding on the preprocessed dispatching business natural language text (dispatch business natural language text refers to professional documents and text data written and communicated by dispatchers or maintenance personnel in natural language during power grid dispatching operations, used for applying for, instructing, recording, and standardizing switching operations and related work, including maintenance tickets, switching operation instruction tickets, equipment commissioning sheets, dispatching procedures, equipment models, new equipment startup plans, and arrangement notification documents) to obtain an input vector sequence. The input vector sequence is then encoded using a natural language parsing model to obtain an information encoding vector. This information encoding vector is mapped to sequence labels conforming to the BIO annotation pattern. Based on these sequence labels, identification and... This design extracts operational equipment, equipment status, operational events, operational rules, operational event intent, and operational logic information from the natural language text of scheduling operations, resulting in structured objectified data. Through BIO sequence labeling, the natural language parsing model can accurately classify each word in a sentence into a predefined category (such as equipment, status, or event), thereby parsing out various implicit information elements in the text. Based on sequence labels, scattered words are recombined into complete entities and assembled into key-value pairs of objectified data. This transforms the instructions written by dispatchers and understood through human experience into standardized data that can be accurately recognized and processed by computers. The safety verification module is used to perform topology analysis (confirming electrical connections), graphic comparison (verifying the on-site status of operating equipment), intelligent ticketing, maintenance linkage, sand table simulation (simulating risks by calculating power flow), and pre-order verification based on the objectified data before the switching operation is executed. During the switching operation, the module verifies key links in the operation instruction ticket (preventing accidental entry into the interval), verifies the status of operating equipment (mandating secondary confirmation), and verifies the timing of operation events (optimizing the order of operation events) based on the objectified data. This design, through objectified data, can transform safety verification from relying on the dispatcher's personal experience and post-event review to an automated and real-time safety verification embedded in every step of the switching operation process, based on data and rules.

[0018] In the above technical solution, the security verification module is also used to perform a post-evaluation of the entire process of operations (drafting, reviewing, issuing orders, issuing orders, and archiving) from before the switching operation to the completion of the switching operation based on the objectified data after the switching operation is completed, thereby establishing a full-process control mechanism. The above design establishes a data-driven closed loop from evaluation to feedback to optimization, which can realize the qualitative change of switching operation security verification from single transaction processing to continuous closed-loop optimization, and build a full-process intelligent control system that can learn and improve itself.

[0019] In the above technical solution, the specific process of constructing the natural language parsing model is as follows: The knowledge corpus in the field of power grid dispatching includes historical maintenance tickets, switching operation instruction tickets, equipment commissioning sheets, dispatching procedures, and equipment models; Based on a general pre-trained model based on the Transformer architecture, a masking language model is used to randomly mask some words in the text of the power grid dispatching domain knowledge corpus. The general pre-trained model based on the Transformer architecture then predicts the masked words, thereby learning the semantics of the words in the text context. The next sentence prediction task allows the general pre-trained model based on the Transformer architecture to determine whether two sentences in the text of the power grid dispatching domain knowledge corpus are in a continuous relationship, thereby generating a basic language model. A dictionary adapter is inserted between the Transformer layers of the basic language model. When the text of the power grid dispatch domain knowledge corpus passes through the basic language model, the dictionary adapter matches the currently processed character vector with the word vector in the power grid dispatch professional feature dictionary, and performs domain-adaptive training through attention mechanism fusion to enhance the feature representation of domain-related words and obtain an enhanced domain language model. The enhanced domain language model yields a vector sequence of text from the power grid dispatching domain knowledge corpus. This vector sequence is then input into a bidirectional long short-term memory network to capture long-distance dependencies and sequential relationships, ultimately resulting in a natural language parsing model. This design is based on a general pre-trained model with a Transformer architecture. Self-supervised pre-training internalizes power grid dispatching industry knowledge, a dictionary adapter enhances key information, and a bidirectional long short-term memory network captures sequential dependencies. This results in a natural language parsing model with powerful semantic understanding and domain knowledge, capable of accurately understanding the complex semantics and operational logic of dispatching business natural language text.

[0020] In the above technical solution, the specific process of performing word embedding and positional encoding on the preprocessed scheduling business natural language text to obtain the input vector sequence is as follows: The scheduling business natural language text is preprocessed (the text data is split into segments, truncated into paragraphs, deduplicated, and formatted with serial numbers and numbers to obtain the preprocessed scheduling business natural language text). Based on predefined rules conforming to the BIO annotation pattern, each character in the preprocessed scheduling business natural language text is assigned a BIO tag, where B represents the first character of an entity word, I represents characters other than the first character of an entity word, and O represents characters that do not belong to any entity word (e.g., "terminal box and line side protection"). In the sentence "After changing the startup scheme", the word "terminal" in "terminal box" is marked as "B-device", "sub" and "box" are marked as "I-device", similarly, "line" is marked as "B-line", "road" is marked as "I-line", and other words are marked as "O". "[CLS]" and "[SEP]" are inserted at the beginning and end of the sentence respectively. Through word embedding technology, each word in the preprocessed scheduling business natural language text that has been marked with BIO tags is converted into a vector, and a position encoding vector is added to obtain the input vector sequence. If the preprocessed scheduling business natural language text contains sentence L, and sentence L has n characters (i.e., a length of n), then by using word embedding technology to split each character of the sentence and mapping each character to a vector representation, then... ; in Represents the nth word; encoded by positional vector. Indicates the position of corresponding words in a sentence, where, The vector represents the position of the nth character; the input vector sequence is obtained through the vector representation of sentence L and the corresponding position encoding vector; the above design transforms unstructured scheduling business natural language text into a standardized numerical representation rich in semantic and positional information, which can provide a foundation for deep parsing of natural language parsing models.

[0021] In the above technical solution, the process of encoding the input vector sequence using a natural language parsing model to obtain an information encoding vector, mapping the information encoding vector to a sequence label conforming to the BIO annotation pattern, and identifying and extracting the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information from the natural language text of the scheduling business based on the sequence label, to obtain structured objectified data, is as follows: The input vector sequence is forward computed using the Transformer encoder in a natural language parsing model to obtain the output vector T after Transformer encoding. ; The output vector T is forward computed using a bidirectional long short-term memory (BSSM) network in a natural language parsing model to obtain the information encoding vector H after encoding by the BSSM network. , h n The vector representing the position of the nth character in the sentence; The information encoding vector H is computed forward using a multilayer perceptron classification layer (MLP classification layer) in a natural language processing model. The Softmax function is then used to calculate the BIO label probability distribution for each word in the information encoding vector H (BIO label probability represents the likelihood of a word belonging to each BIO label). The BIO label with the highest probability is selected as the prediction result for that word, thus mapping the information encoding vector H to sequence labels conforming to the BIO labeling pattern. Based on the sequence labels, the original text at the corresponding positions is extracted, thereby identifying and extracting the operating devices, operating device status, operating events, operating rule information, operating event intent, and operating logic information from the scheduling business natural language text. This information is then assembled into structured object-oriented data in key-value pair format. This design can extract key business elements from scheduling business natural language text, not only identifying isolated entities in the text but also reconstructing complete operating event intent and operating logic information through the relationships between sequence labels, thus outputting standard key-value pair object-oriented data.

[0022] In the above technical solution, the specific process of performing topology analysis, image-text comparison, intelligent ticketing, maintenance linkage, sand table simulation, and pre-order verification based on the objectified data before the switching operation is executed is as follows: Topology analysis is performed based on the objectified data. According to the ID of the operating device in the objectified data, it can be queried in the database of the SCADA system. By utilizing the association relationship between topology point numbers ("sequential connection, busbar cut-off" principle), the operating device is automatically located and the topology point number associated with the operating device is obtained. Then, all electrically directly or indirectly related primary devices (such as circuit breakers, disconnect switches, grounding switches and busbars) are found, and the electrical connection relationship between operating devices is automatically obtained. Based on the objectified data, a graphic comparison is performed. According to the ID of the operating device, the status of the operating device, and the intent of the operating event in the objectified data, the topology analysis function is invoked to generate a current status list of all main devices and their fully associated devices involved in the operation instruction ticket. The status of the operating devices in the list is compared with the actual status of the operating devices on site, and the status of the operating devices is confirmed or anomalies are reported to ensure the accuracy of the status of the operating devices before the ticket is drafted. Based on the objectified data, intelligent ticket generation is performed. After the dispatching and operation personnel draft the operation tasks and method requirements, the operation tasks and method requirements are parsed using a natural language parsing model to extract the operation equipment and operation event intent. The intelligent ticket generation model, which is pre-trained based on a large number of historical typical operation tickets and dispatching procedures, is called to automatically generate a complete operation instruction ticket that conforms to the dispatching specifications and is logically correct, according to the extracted operation equipment and operation event intent. Based on the objectified data, maintenance linkage is carried out. All maintenance tickets to be executed are periodically retrieved from the maintenance ticket system (maintenance tickets are standardized work certificates and management processes used in the power system to apply for, approve and record equipment shutdown for maintenance, testing or maintenance work). Natural language parsing model is used to determine whether each maintenance ticket needs to be written with a corresponding operation instruction ticket and the time window for writing the ticket. When it is detected that the dispatching operator has missed writing the operation instruction ticket within the time window, a pop-up window and a voice alarm based on speech synthesis technology are automatically triggered to remind the operator. Based on the objectified data, a sand table simulation is performed. According to the objectified data of the operation instruction ticket, the changes in the status of the operating equipment are dynamically simulated (the primary wiring diagram of the substation can be visualized based on the D5000 platform, and the entire process of the operation instruction ticket execution can be dynamically simulated). Power flow calculation is automatically triggered to analyze the power grid status after the simulated changes in the status of the operating equipment (such as whether the line load and voltage exceed the limit). Based on the objectified data, pre-order verification is performed. For power outage operations, the verification checks whether the objectified data contains key requirements (such as "check that a certain device has been energized before operation"), and verifies whether the associated maintenance ticket has been completed, the associated safety and stabilization measures have been started, and whether the operation of the superior-approved equipment has been authorized. If any condition is not met, the operation is prohibited. For power restoration operations, the verification checks whether the associated maintenance ticket has been completed, the associated protection setting sheet has been executed, and the associated power restoration commissioning sheet or new equipment commissioning plan has been started. If any condition is not met, the operation is prohibited. The above design establishes a complete preventive safety protection system through pre-operation safety verification of switching operations, which can reduce the risk of human error, improve the safety and reliability of switching operations, and improve operational efficiency through automation.

[0023] In the above technical solution, the specific process of performing key link verification, equipment status verification, and timing verification based on the objectified data during the switching operation is as follows: Based on the objectified data, key steps are verified to identify critical operational steps in the operation instruction ticket (such as line charging operation and operation of equipment authorized by the superior). When a critical operational step is simulated or actually executed, an enhanced verification mechanism is automatically triggered: For charging operations, a secondary confirmation request (including pop-up windows and voice broadcasts) is sent to the operating units on both sides of the line. The operation instruction is only allowed to be issued after confirmation is completed on both sides; For operation of equipment authorized by the superior, the shift leader authorization mechanism is triggered, and an authorization request interface pops up. The operation instruction can only continue to be executed after the shift leader confirms the authorization. Based on the objectified data, the device status is checked. When an operation command is executed, the operation device and its status are checked in the objectified data, and the actual status of the corresponding operation device is queried in real time. If the operation device status in the objectified data is inconsistent with the actual status of the operation device (e.g., the operation device status command in the objectified data requires "disconnect" but the actual status of the operation device is already "disconnected"), the command is prohibited and an alarm is triggered. Based on the objectified data, timing verification is performed. When there are multiple operation tasks, the timing requirements are parsed from the approval opinions of the associated maintenance tickets. If there are no explicit requirements, power flow calculation simulation is automatically performed to simulate all possible operation sequences and select the optimal timing sequence with the least impact on the power grid. The above design establishes a real-time, dynamic, and multi-layered intelligent safety protection system during the execution of switching operations, thereby improving the reliability and safety of switching operations.

[0024] In the above technical solution, after the switching operation is completed, a post-evaluation is performed on the entire process from before the switching operation to the completion of the switching operation based on the objectified data, thereby establishing a full-process control mechanism. The specific process is as follows: Based on the objectified data, all objectified data, operation logs, personnel behavior records, timestamps, operating equipment status change sequences and verification results generated in the entire process (drafting, reviewing, issuing orders, issuing orders and archiving) from the start of the switching operation to its completion are recorded to form an electronic operation archive. Based on the electronic operation archive and the objectified data, a multi-dimensional post-evaluation analysis is performed to obtain the post-evaluation results. Based on the post-evaluation results, typical error cases, high-risk operations, and weak links are selected for assessment and training of dispatching and operation personnel. New typical operation patterns, optimized operation logic, and common types of misoperations discovered in the post-evaluation results are used as feedback data to incrementally learn and iteratively optimize the knowledge base of the intelligent ticketing model and natural language parsing model. By mining historical operation data, potential systemic risk points and rule loopholes are identified, and optimization suggestions are automatically generated and the rule base update process is triggered, thereby achieving comprehensive intelligent control of switching operations. This design establishes a closed-loop learning system from evaluation to feedback to optimization through full-process datafication, multi-dimensional post-evaluation, and two-way feedback optimization, transforming switching operation safety management from a static, experience-dependent model into a dynamic, self-improving intelligent control system.

[0025] Example 2 A safety verification method for switching operations based on artificial intelligence, such as... Figure 2 As shown, an enhanced domain language model is obtained using self-supervised pre-training technology and a power grid dispatching professional feature dictionary. This model is then combined with a bidirectional long short-term memory network to obtain a natural language parsing model. The natural language parsing model is used to transform the preprocessed dispatching business natural language text into structured objectified data that identifies operating equipment, operating equipment status, operating events, operating rule information, operating event intent, and operating logic information. Based on the objectified data, topology analysis, image-text comparison, intelligent ticketing, maintenance linkage, sand table simulation, and pre-order verification are performed before the switching operation is executed. During the switching operation, key link verification, operating equipment status verification, and operating event timing verification are performed.

[0026] The specific methods for safety verification of switching operations include the following steps: Self-supervised pre-training of knowledge corpus in the power grid dispatching domain is performed using a masked language model and a next sentence prediction task to generate a basic language model. The basic language model is then subjected to domain-adaptive training using a power grid dispatching professional feature dictionary to obtain an enhanced domain language model. Finally, the enhanced domain language model is combined with a bidirectional long short-term memory network to obtain a natural language parsing model. Word embedding and position encoding are performed on the preprocessed scheduling business natural language text to obtain an input vector sequence. The input vector sequence is then encoded using a natural language parsing model to obtain an information encoding vector. The information encoding vector is mapped to a sequence label that conforms to the BIO annotation pattern. Based on the sequence label, the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information in the scheduling business natural language text are identified and extracted to obtain structured objectified data. Before the switching operation is executed, topology analysis, image and text comparison, intelligent ticketing, maintenance linkage, sand table simulation and pre-order verification are performed based on the objectified data; during the switching operation, key links in the operation instruction ticket, operation equipment status verification and operation event timing verification are performed based on the objectified data.

[0027] Example 3 A computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 2.

[0028] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A safety verification system for switching operations based on artificial intelligence, characterized in that, It includes: The model building module is used to perform self-supervised pre-training on the knowledge corpus of power grid dispatching through masked language model and next sentence prediction task to generate basic language model. The basic language model is then subjected to domain-adaptive training through power grid dispatching professional feature dictionary to obtain enhanced domain language model. The enhanced domain language model is then combined with bidirectional long short-term memory network to obtain natural language parsing model. The text parsing module is used to perform word embedding and position encoding on the preprocessed scheduling business natural language text to obtain an input vector sequence. The input vector sequence is then encoded by a natural language parsing model to obtain an information encoding vector. The information encoding vector is mapped to a sequence label that conforms to the BIO labeling pattern. Based on the sequence label, the module identifies and extracts the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information from the scheduling business natural language text to obtain structured objectified data. The safety verification module is used to perform topology analysis, image and text comparison, intelligent ticketing, maintenance linkage, sand table simulation and pre-order verification based on the objectified data before the switching operation is executed; During the switching operation, the key links in the operation instruction ticket, the status of the operating equipment, and the timing of the operation events are checked based on the objectified data.

2. The artificial intelligence-based switching operation safety verification system according to claim 1, characterized in that: The safety verification module is also used to perform a post-evaluation of the entire process from before the switching operation to the completion of the switching operation based on the objectified data after the switching operation is completed, thereby establishing a full-process control mechanism.

3. The artificial intelligence-based switching operation safety verification system according to claim 1, characterized in that: The specific process of constructing the natural language parsing model is as follows: The knowledge corpus in the field of power grid dispatching includes historical maintenance tickets, switching operation instruction tickets, equipment commissioning sheets, dispatching procedures, and equipment models; Based on a general pre-trained model based on the Transformer architecture, a masking language model is used to randomly mask some words in the text of the power grid dispatching domain knowledge corpus. The general pre-trained model based on the Transformer architecture then predicts the masked words, thereby learning the semantics of the words in the text context. The next sentence prediction task allows the general pre-trained model based on the Transformer architecture to determine whether two sentences in the text of the power grid dispatching domain knowledge corpus are in a continuous relationship, thereby generating a basic language model. A dictionary adapter is inserted between the Transformer layers of the basic language model. When the text of the power grid dispatch domain knowledge corpus passes through the basic language model, the dictionary adapter matches the currently processed character vector with the word vector in the power grid dispatch professional feature dictionary, and performs domain-adaptive training through attention mechanism fusion to enhance the feature representation of domain-related words and obtain an enhanced domain language model. The enhanced domain language model is used to obtain the text vector sequence of power grid dispatch domain knowledge corpus. The vector sequence is then input into a bidirectional long short-term memory network to capture long-distance dependencies and sequential relationships in the vector sequence, and finally a natural language parsing model is obtained.

4. The switching operation safety verification system based on artificial intelligence according to claim 1, characterized in that: The specific process of performing word embedding and positional encoding on the preprocessed scheduling business natural language text to obtain the input vector sequence is as follows: The scheduling business natural language text is preprocessed. According to the predefined rules that conform to the BIO annotation pattern, each character in the preprocessed scheduling business natural language text is assigned a BIO tag. B represents the first character in the sentence that belongs to an entity word, I represents the character in the sentence that does not belong to any entity word, and O represents the character that does not belong to any entity word. Each character in the preprocessed scheduling business natural language text that has been annotated with BIO tags is converted into a vector through word embedding technology, and a position encoding vector is added to obtain the input vector sequence. If the preprocessed scheduling business natural language text contains sentence L, and sentence L has n characters (i.e., a length of n), then by using word embedding technology to split each character of the sentence and mapping each character to a vector representation, then... ; in Represents the nth word; encoded by positional vector. Indicates the position of corresponding words in a sentence, where, The vector representing the position of the nth character; the input vector sequence is obtained by combining the vector representation of sentence L with the corresponding positional encoding vector.

5. The artificial intelligence-based switching operation safety verification system according to claim 1, characterized in that: The process of encoding an input vector sequence using a natural language processing model to obtain an information encoding vector, mapping the information encoding vector to sequence labels conforming to the BIO annotation pattern, and identifying and extracting operational devices, operational device status, operational events, operational rule information, operational event intent, and operational logic information from the natural language text of scheduling services based on these sequence labels to obtain structured objectified data is as follows: The input vector sequence is forward computed using the Transformer encoder in a natural language parsing model to obtain the output vector T after Transformer encoding. ; The output vector T is forward computed using a bidirectional long short-term memory (BSSM) network in a natural language parsing model to obtain the information encoding vector H after being encoded by the BSSM network. , h n The vector representing the position of the nth character in the sentence; The information encoding vector H is computed forward by the multilayer perceptron classification layer in the natural language parsing model. The probability distribution of BIO labels for each word in the information encoding vector H is calculated by combining the Softmax function. The BIO label with the highest probability is selected as the prediction result for that word, thus mapping the information encoding vector H into a sequence label that conforms to the BIO labeling pattern. Based on the sequence label, the original text at the corresponding position is extracted, thereby identifying and extracting the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information in the natural language text of the scheduling business, and assembling them into structured object data in the form of key-value pairs.

6. The switching operation safety verification system based on artificial intelligence according to claim 1, characterized in that: Before the switching operation is executed, the specific process of performing topology analysis, image-text comparison, intelligent ticketing, maintenance linkage, sand table simulation, and pre-order verification based on the objectified data is as follows: Topology analysis is performed based on the objectified data. According to the ID of the operating device in the objectified data, it can be queried in the database of the SCADA system. By utilizing the association between topology point numbers, the operating device is automatically located and the topology point number associated with the operating device is obtained. Then, all electrically directly or indirectly related primary devices are found, and the electrical connection relationship between the operating devices is automatically obtained. Based on the objectified data, a graphic comparison is performed. According to the ID of the operating device, the status of the operating device, and the intent of the operating event in the objectified data, the topology analysis function is invoked to generate a current status list of all main devices and their fully associated devices involved in the operation instruction ticket. The status of the operating devices in the list is compared with the actual status of the operating devices on site, and the status of the operating devices is confirmed or anomalies are reported to ensure the accuracy of the status of the operating devices before the ticket is drafted. Intelligent ticketing is performed based on the objectified data. After the scheduling and operation personnel draft the operation tasks and method requirements, the operation tasks and method requirements are parsed using a natural language parsing model to extract the operation equipment and operation event intent. The system invokes a pre-trained intelligent ticket generation model based on a large number of historical typical operation tickets and scheduling procedures to automatically generate complete operation instruction tickets that conform to scheduling specifications and are logically correct, according to the extracted operation equipment and operation event intent. Based on the objectified data, maintenance linkage is carried out. All maintenance tickets to be executed are periodically obtained from the maintenance ticket system. Natural language parsing model is used to determine whether each maintenance ticket needs to be written with corresponding operation instruction tickets and the time window for writing the tickets. When it is detected that the dispatching operator has missed writing the operation instruction ticket within the time window, a pop-up window and a voice alarm based on speech synthesis technology are automatically triggered to remind the operator. Based on the objectified data, a sand table simulation is performed. According to the objectified data of the operation instruction ticket, the state changes of the operating equipment are dynamically simulated, and power flow calculation is automatically triggered to analyze the power grid state after the simulated state changes of the operating equipment. Pre-order verification is performed based on the objectified data. For power outage operations, the verification checks whether the objectified data contains key requirements and verifies whether the associated maintenance ticket has been completed, the associated safety and stability measures have been started, and whether the operation of the superior-approved equipment has been authorized. If any condition is not met, the operation is prohibited. For power restoration operations, the verification checks whether the associated maintenance ticket has been completed, the associated protection setting sheet has been executed, and the associated power restoration commissioning sheet or new equipment commissioning plan has been started. If any condition is not met, the operation is prohibited.

7. The switching operation safety verification system based on artificial intelligence according to claim 1, characterized in that: During the switching operation, the specific process of performing key link verification, equipment status verification, and timing verification based on the objectified data is as follows: Based on the objectified data, key steps are verified to identify key operational steps in the operation instruction ticket. When a key operational step is simulated or actually executed, an enhanced verification mechanism is automatically triggered: For charging operations, a secondary confirmation request is sent to the operating units on both sides of the line. The operation instruction is only allowed to be issued after both sides have completed the confirmation; For operations of equipment authorized by the superior, the shift leader authorization mechanism is triggered, and an authorization request interface pops up. The operation instruction can only continue to be executed after the shift leader confirms the authorization. Based on the objectified data, the device status is checked. When an operation command is executed, the actual status of the corresponding operation device is queried in real time according to the operation device and operation device status in the objectified data. If the operation device status in the objectified data is inconsistent with the actual status of the operation device, the command is prohibited and an alarm is triggered. Based on the objectified data, timing verification is performed. When there are multiple operation tasks, timing requirements are parsed from the approval opinions of the associated maintenance tickets. If there are no explicit requirements, power flow calculation simulation is automatically performed to simulate all possible operation sequences and select the optimal timing sequence with the least impact on the power grid.

8. The switching operation safety verification system based on artificial intelligence according to claim 2, characterized in that: After the switching operation is completed, a post-evaluation of the entire process from before the switching operation to the completion of the switching operation is performed based on the objectified data, thereby establishing a full-process control mechanism. The specific process is as follows: Based on the objectified data, all objectified data, operation logs, personnel behavior records, timestamps, operating equipment status change sequences and verification results generated in the entire process from before the switching operation to the completion of the switching operation are recorded to form an electronic operation archive. Based on the electronic operation archive and the objectified data, a multi-dimensional post-evaluation analysis is performed to obtain the post-evaluation results. Based on the post-evaluation results, typical error cases, high-risk operations, and weak links are selected for use in the assessment and training of dispatching and operation personnel. The new typical operating patterns, optimized operating logic, and common types of misoperations discovered in the post-evaluation results are used as feedback data to incrementally learn and iteratively optimize the knowledge base of the intelligent ticketing model and the natural language parsing model. By mining historical operating data, potential systemic risk points and rule loopholes are identified, optimization suggestions are automatically generated, and the rule base update process is triggered, thereby achieving comprehensive intelligent control over the switching operation business.

9. A safety verification method for switching operations based on artificial intelligence, characterized in that, It includes the following steps: Self-supervised pre-training of knowledge corpus in the power grid dispatching domain is performed using a masked language model and a next sentence prediction task to generate a basic language model. The basic language model is then subjected to domain-adaptive training using a power grid dispatching professional feature dictionary to obtain an enhanced domain language model. Finally, the enhanced domain language model is combined with a bidirectional long short-term memory network to obtain a natural language parsing model. Word embedding and position encoding are performed on the preprocessed scheduling business natural language text to obtain an input vector sequence. The input vector sequence is then encoded using a natural language parsing model to obtain an information encoding vector. The information encoding vector is mapped to a sequence label that conforms to the BIO annotation pattern. Based on the sequence label, the operation device, operation device status, operation event, operation rule information, operation event intent, and operation logic information in the scheduling business natural language text are identified and extracted to obtain structured objectified data. Before the switching operation is executed, topology analysis, image and text comparison, intelligent ticketing, maintenance linkage, sand table simulation and pre-order verification are performed based on the objectified data. During the switching operation, the key links in the operation instruction ticket, the status of the operating equipment, and the timing of the operation events are checked based on the objectified data.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of claim 9.

Images