Operation ticket text auditing method and device, equipment and storage medium

Abstract

A method, device and equipment for auditing operation ticket text and a computer readable storage medium, comprising: inputting a procedure text of an obtained to-be-audited operation ticket text into a first deep learning model to obtain a structured auditing rule output by the first deep learning model; checking the to-be-audited operation ticket according to the structured auditing rule to obtain a rule checking result; inputting the obtained to-be-audited operation ticket text into a second deep learning model to obtain a historical matching result output by the second deep learning model; and generating an auditing conclusion report according to the rule checking result and the historical matching result, thereby solving the technical problems of a traditional auditing system in related technologies, such as dependence on static rules and low data utilization, and achieving the effects of breaking through the limitation of a static rule library and improving data utilization.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, specifically to a method, apparatus, device, and computer-readable storage medium for reviewing operation ticket text. Background Technology

[0002] Switching operations of electrical equipment are a core task of power grid operation and maintenance. Operation tickets, as a crucial control mechanism for regulating operational procedures, mitigating the risk of misoperation, and ensuring equipment and personnel safety, require rigorous compliance review, which is a key aspect of power safety production control. With the continuous expansion of the power grid architecture, the increasing number of substations, and the diversification of electrical equipment types and operating modes, the number of daily operation tickets issued has increased significantly. The traditional review model, combining manual and intelligent assistance, is gradually becoming insufficient to meet the current requirements for efficient and precise safety control.

[0003] Currently, most mainstream operation ticket approval systems adopt a static rule base verification model. The approval logic relies entirely on the fixed settings of general standards such as the "Power Safety Work Regulations," resulting in lagging rule system updates and insufficient flexibility. Different regions and substations have different customized operation procedures due to equipment modifications, wiring methods, and on-site maintenance habits. Fixed static rules cannot achieve adaptive matching for personalized scenarios, easily leading to problems such as rule adaptation failure, misjudgment, and missed judgment. The universality of the approval process and its practicality in the field are poor.

[0004] Meanwhile, the power industry has accumulated massive amounts of historical compliant operation ticket data over the years, covering standard operating logic under various operating conditions and equipment scenarios, which has extremely high reference value. However, existing auditing technologies lack data mining and intelligent analysis capabilities, fail to structure and reuse historical operation samples, and cannot assist intelligent auditing through operation content similarity comparison, classic case reference, or other methods. A large amount of high-quality historical data resources are idle and wasted, and the auditing work relies excessively on hard rule validation and manual experience review, resulting in a low overall level of intelligence, low auditing efficiency, and significant risks of human error, making it difficult to meet the development needs of modern power grid safety and lean operation and maintenance. Summary of the Invention

[0005] This application provides a method, apparatus, device, and computer-readable storage medium for reviewing operation ticket text, which can solve the technical problems of traditional review systems relying on static rules and having low data utilization in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for reviewing operation ticket text, including: The procedure text of the operation ticket to be reviewed is input into the first deep learning model to obtain the structured review rules output by the first deep learning model. The first deep learning model is generated by training using power procedure documents as training data and has the ability to extract rule logic from text. The operation ticket to be reviewed is verified according to the structured review rules, and the rule verification result is obtained; The obtained text of the operation ticket to be reviewed is input into the second deep learning model to obtain the historical matching results output by the second deep learning model. The second deep learning model is generated by training using a historical compliant operation ticket database as training data and has the ability to obtain the similarity of historical operation tickets. Based on the rule verification results and the historical matching results, an audit conclusion report is generated.

[0007] In conjunction with the first aspect, in one implementation, the step of inputting the procedure text of the obtained operation ticket text to be reviewed into a first deep learning model to obtain the structured review rules output by the first deep learning model includes: The procedure text of the obtained operation ticket to be reviewed is input into the first deep learning model; The encoder in the first deep learning model extracts professional entity information from the procedure text; The professional entity information is semantically parsed and information is extracted by the decoder in the first deep learning model to generate an intermediate logical representation. The intermediate logic representation format is converted by the rule compiler in the first deep learning model to obtain the structured audit rules output by the first deep learning model.

[0008] In conjunction with the first aspect, in one implementation, the step of verifying the operation ticket to be reviewed according to the structured review rules and obtaining the rule verification result includes: The operation ticket to be reviewed is subjected to semantic structured parsing to extract entity information and generate key operation elements, wherein the entity information includes step sequence, device object and state verb; The key operational elements and the structured audit rules are judged by the pre-set rule verification engine to obtain the rule verification result, wherein the rule verification result includes pass and compliance information or fail and violation information.

[0009] In conjunction with the first aspect, in one implementation, the step of inputting the obtained text of the operation ticket to be reviewed into a second deep learning model to obtain the historical matching results output by the second deep learning model includes: The text of the operation ticket to be reviewed is input into the second deep learning model; The semantic representation vector is obtained by semantically encoding the operation ticket to be reviewed through the BERT pre-trained layer in the second deep learning model; Temporal features are extracted from the semantic representation vector using the BiLSTM sequence modeling layer in the second deep learning model to obtain a sequence feature vector; By focusing key features on the sequence feature vector through the multi-head self-attention layer in the second deep learning model, a fixed-dimensional operation ticket feature vector is obtained. The retrieval and matching layer in the second deep learning model matches historical feature vectors based on the operation ticket feature vector, and calculates similarity scores using a cosine similarity and edit distance fusion algorithm to obtain historical matching results output by the second deep learning model. The historical matching results include similarity scores and difference information.

[0010] In conjunction with the first aspect, in one implementation, the rule verification result includes a verification pass status and corresponding compliance information, or a verification fail status and corresponding violation information; the historical matching result includes a similarity score and difference information; generating an audit conclusion report based on the rule verification result and the historical matching result includes: The similarity score is compared with a preset reliability threshold; When the rule verification result is a verification failure, the final review conclusion is determined to be a review failure, and the violation information is written into the review conclusion report; When the rule verification result is "verification passed" and the similarity score is greater than or equal to the preset reliability threshold, the final review conclusion is determined to be "review passed". When the rule verification result is in the verification passed state and the similarity score is less than the preset reliability threshold, the final review conclusion is determined according to the preset conflict resolution strategy, which includes the rule priority strategy. Generate an audit conclusion report that includes the final audit conclusion, the compliance information or the violation information, and the discrepancy information.

[0011] In conjunction with the first aspect, in one implementation, the first deep learning model is generated by training using power regulation documents as training data, including: Obtain first training set data, wherein the first training set data includes multiple sets of first training data, and each set of first training data includes power regulations clauses and corresponding labeled structured rules; The first pre-set neural network model is trained based on the first training set data to generate a first deep learning model, wherein the first pre-set neural network model is DeBERTa-V3.

[0012] In conjunction with the first aspect, in one implementation, the second deep learning model is generated by training using a historical compliant operation ticket database as training data, including: Obtain second training set data, wherein the second training set data includes multiple sets of second training data, and each set of second training data includes a first historical operation ticket, a second historical operation ticket, and labeled similarity and difference information of the first historical operation ticket and the second historical operation ticket; The second pre-set neural network model is trained based on the second training set data to generate a second deep learning model. The second pre-set neural network model is a composite neural network using a BERT pre-training layer, a BiLSTM sequence modeling layer, and a multi-head self-attention layer.

[0013] Secondly, embodiments of this application provide a device for reviewing operation ticket text, the device comprising: The first acquisition module is used to input the procedure text of the acquired operation ticket text to be reviewed into the first deep learning model to obtain the structured review rules output by the first deep learning model. The first deep learning model is generated by training using power procedure documents as training data and has the ability to extract rule logic from text. The second acquisition module is used to verify the operation ticket to be reviewed according to the structured review rules and obtain the rule verification result; The third acquisition module is used to input the acquired text of the operation ticket to be reviewed into the second deep learning model to obtain the historical matching results output by the second deep learning model. The second deep learning model is generated by training using a historical compliant operation ticket database as training data and has the ability to obtain the similarity of historical operation tickets. The generation module is used to generate an audit conclusion report based on the rule verification results and the historical matching results.

[0014] Thirdly, embodiments of this application provide an operation ticket text review device, which includes a processor, a memory, and an operation ticket text review program stored in the memory and executable by the processor. When the operation ticket text review program is executed by the processor, it implements the steps of the operation ticket text review method described above.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing an operation ticket text review program, wherein when the operation ticket text review program is executed by a processor, the steps of the operation ticket text review method described above are implemented.

[0016] The beneficial effects of the technical solutions provided in this application include: By inputting the procedural text of the obtained operation ticket to be reviewed into a first deep learning model, structured review rules are obtained from the output of the first deep learning model. The first deep learning model is trained using power regulations as training data and has the ability to extract rule logic from text. The operation ticket to be reviewed is then validated according to the structured review rules, and rule validation results are obtained. Next, the obtained operation ticket text is input into a second deep learning model, and historical matching results are obtained from the output of the second deep learning model. The second deep learning model is trained using a historical compliant operation ticket database as training data and has the ability to obtain historical operation ticket similarity. Based on the rule validation results and the historical matching results, an audit conclusion report is generated. This approach solves the technical problems of traditional audit systems relying on static rules and having low data utilization in related technologies, achieving the effect of overcoming the limitations of static rule databases and improving data utilization. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the method for reviewing the operation ticket text in this application; Figure 2 This is a schematic diagram of the functional modules of an embodiment of the device for reviewing operation ticket text in this application; Figure 3 This is a schematic diagram of the hardware structure of the device for reviewing operation ticket text involved in the embodiments of this application. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0019] First, some of the technical terms used in this application will be explained to help those skilled in the art understand this application.

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0021] Firstly, embodiments of this application provide a method for reviewing operation ticket text.

[0022] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the method for reviewing the operation ticket text in this application. Figure 1 As shown, the methods for reviewing operation ticket texts include: Step S10: Input the procedure text of the obtained operation ticket text to be reviewed into the first deep learning model to obtain the structured review rules output by the first deep learning model. The first deep learning model is generated by training using power procedure documents as training data and has the ability to extract rule logic from text. As an example, the procedural text of the operation ticket to be reviewed is input into the first deep learning model; the encoder in the first deep learning model extracts professional entity information from the procedural text; the decoder in the first deep learning model performs semantic parsing and information extraction on the professional entity information to generate an intermediate logical representation; the rule compiler in the first deep learning model converts the format of the intermediate logical representation to obtain the structured review rules output by the first deep learning model. For example, the system retrieves procedural text related to the current operation ticket to be reviewed from the power regulations knowledge base. The procedural text includes, but is not limited to, the "Power Safety Work Regulations," specific operating regulations for power plants and stations, and documents on accident prevention measures. In specific implementation, the retrieved procedural text is first preprocessed, including removing irrelevant characters, paragraph segmentation, and sentence tokenization. The preprocessed procedural text sequence is then used as an input vector and input into the first deep learning model (i.e., the procedural text parsing and rule generation model, Model_R). Example: Input text: "When the circuit breaker is in the open state, operation of the corresponding disconnecting switch is prohibited." The first deep learning model, based on encoder-based professional entity information extraction, includes an encoder module and embeds a power industry professional entity recognition layer at the encoder's front end. When the procedure text enters the encoder, it first passes through the power industry professional entity recognition layer, using a pre-trained power industry named entity recognition (NER) model to identify key professional entities in the text. The professional entity information includes at least equipment objects (such as "circuit breaker" and "disconnecting switch"), equipment states (such as "open state" and "closed state"), and operation actions (such as "operation prohibited" and "operation permitted"). The encoder, based on the Transformer architecture, converts the text sequence containing entity information into a context-sensitive semantic vector representation. Example: The identified entity information is: Equipment object 1 = "circuit breaker", state 1 = "open". The operation action is "Disable operation", and device object 2 is "Isolation switch". The decoder, based on the semantic parsing and intermediate logical representation generation model, receives the semantic vector output by the encoder and performs semantic parsing and information extraction. The decoder focuses on the extracted specialized entity information through an attention mechanism, transforming it into a computer-understandable intermediate logical representation (ILR). ILR is a structured logical expression between natural language and executable code, typically using predicate logic or triples. The rule compiler reads the generated intermediate logical representation (ILR) and converts it into a structured audit rule format that the target rule engine can load. This structured audit rule format can be a JSON tree, an XML file, or a specific rule script (such as Drools rules).Finally, the model outputs the structured audit rules for use by the subsequent rule validation engine.

[0023] The system automatically transforms unstructured power regulations into structured audit rules. Specifically, the entity recognition layer at the encoder's front end addresses the issue of general models failing to accurately identify power-related technical terms; the introduction of Intermediate Logic Representation (ILR) reduces the error rate of generating rules directly from text and improves rule interpretability; and the rule compiler ensures that the generated rules can be directly executed by downstream engines, enabling dynamic adaptation for regulations updates.

[0024] The first deep learning model was generated using electricity regulation documents as training data, including: Obtain first training set data, wherein the first training set data includes multiple sets of first training data, and each set of first training data includes electrical code clauses and corresponding labeled structured rules; train the first pre-set neural network model based on the first training set data to generate a first deep learning model, wherein the first pre-set neural network model is DeBERTa-V3.

[0025] Step S20: Verify the operation ticket to be reviewed according to the structured review rules and obtain the rule verification result; As an example, the system performs semantic structured parsing on the operation ticket to be reviewed to extract entity information and generate key operation elements. The entity information includes step sequences, equipment objects, and state verbs. A pre-built rule verification engine then judges the key operation elements and structured review rules to obtain rule verification results. These results include pass / fail and compliance information, or fail / violation information. For example, after obtaining the text of the operation ticket to be reviewed, the system calls the semantic parsing module to perform structured processing. The semantic parsing module can employ a neural network structure similar to the encoder of the first deep learning model, or it can use a rule slot-filling technique. Specifically, the system traverses each operation step in the operation ticket, extracting the following three types of entity information to generate key operation elements: 1. Step sequence: Identifies the sequential numbering of operation steps, used to verify the logical order of operations.

[0026] 2. Equipment Object: Identify the specific electrical equipment name and number involved in the operation, such as 101 circuit breaker, 1011 disconnect switch.

[0027] 3. Status Verbs: Identify the actions performed on the equipment and the target state, such as closing, opening, or switching from operation to maintenance. The extracted information is encapsulated into standardized data objects as key operational elements. Example: For the operation ticket text "1. Close 101 circuit breaker", the generated key operational elements are...

[0028] The rule validation engine loading and matching system calls a pre-built rule validation engine to load the structured review rules generated by the first deep learning model. These structured review rules are typically represented as logic trees or condition-action pairs. The rule validation engine substitutes the generated key operational elements one by one into the structured review rules for logical judgment. The judgment process includes: 1. Equipment Matching: Check whether the equipment objects in the key operational elements are within the scope of rule monitoring.

[0029] 2. State Logic Verification: Checks whether the state verbs of the current operation step meet the preconditions and postconditions defined in the rules. For example, the rules may stipulate that before closing a circuit breaker, it must be confirmed that the corresponding disconnecting switch has been closed.

[0030] 3. Sequence Logic Verification: Based on the step sequence, check whether the order of operation steps violates the timing constraints in the rules. Example: If the structured audit rule states that closing the 1011 disconnector is prohibited if the 1011 circuit breaker is not closed, and the engine detects that the current critical operation element is closing the 1011 circuit breaker, and the pre-state check finds that the "1011 disconnector" status is "open," then a logic conflict is determined. Based on the determination, a structured rule verification result is generated. The rule verification result has two states: 1. Verification Approval Status and Compliance Information: When key operational elements meet all matching structured audit rules, an approval status is generated. The compliance information includes the compliant rule ID, compliance step number, and a description of the compliance reason.

[0031] 2. Verification of Failure Status and Violation Information: When a key operational element violates any structured review rule, a failure status is generated. The violation information includes the violated rule ID, the violation step number, the violation type (e.g., incorrect sequence, status conflict), and suggested corrective measures. Through the above embodiments, this invention achieves refined rule verification of operation tickets. Semantic structured parsing transforms unstructured text into elements that can be logically processed by a computer, resolving the ambiguity problem of natural language understanding; the rule verification engine makes judgments based on dynamically generated structured rules, ensuring the real-time nature and accuracy of the review standards; clear pass / fail status and detailed information output provide quantifiable evidence for subsequent decision fusion modules, and also facilitates maintenance personnel in quickly locating the cause of violations, improving the efficiency and security of operation ticket review.

[0032] Step S30: Input the obtained text of the operation ticket to be reviewed into the second deep learning model to obtain the historical matching results output by the second deep learning model. The second deep learning model is generated by training using the historical compliant operation ticket database as training data and has the ability to obtain the similarity of historical operation tickets. As an example, the text of the operation ticket to be reviewed is input into the second deep learning model; the BERT pre-trained layer in the second deep learning model performs semantic encoding on the operation ticket to be reviewed, obtaining a semantic representation vector; the BiLSTM sequence modeling layer in the second deep learning model extracts temporal features from the semantic representation vector, obtaining a sequence feature vector; the multi-head self-attention layer in the second deep learning model focuses on key features of the sequence feature vector, obtaining a fixed-dimensional operation ticket feature vector; the retrieval and matching layer in the second deep learning model matches historical feature vectors based on the operation ticket feature vector, and calculates a similarity score using a cosine similarity and edit distance fusion algorithm, obtaining the historical matching results output by the second deep learning model, where the historical matching results include similarity scores and difference information.

[0033] For example, the text of the operation ticket to be reviewed is input and semantically encoded. The text of the operation ticket to be reviewed is obtained, preprocessed (e.g., word segmentation and denoising), and then input into the second deep learning model. Specifically, the text of the operation ticket to be reviewed is first semantically encoded using a BERT pre-trained layer in the second deep learning model. The BERT model utilizes a bidirectional Transformer encoder structure to capture the contextual dependencies of words in the operation ticket text, outputting a sequence of context vectors containing rich semantic information. Example: For the operation ticket step "Close the 101 circuit breaker," the BERT layer not only encodes the meanings of "close" and "circuit breaker," but also encodes the action-object relationship between them, obtaining the semantic representation vector Vsemantic.

[0034] Temporal feature extraction involves inputting the semantic representation vector into the BiLSTM sequence modeling layer of the second deep learning model. Since the operation tickets have strict temporal order logic (e.g., pulling the isolating switch before pulling the circuit breaker), the BiLSTM (Bidirectional Long Short-Term Memory) layer is used to extract temporal features from the semantic representation vector. The BiLSTM layer uses two LSTM networks, forward and backward, to capture the positive and negative dependencies of the operation step sequence, respectively, outputting a sequence feature vector Vsequence containing temporal information. Example: The BiLSTM layer can identify whether the operation depends on the execution result, thus reflecting the temporal relationship between steps in the vector.

[0035] Key feature focusing and vector generation involve inputting the sequence feature vector into a multi-head self-attention layer in the second deep learning model. This layer focuses on key features of the sequence feature vector. By calculating attention weights for features at different locations, the model automatically assigns higher weights to key operation steps, such as critical equipment operations and safety measure executions, to suppress interference from irrelevant information. After pooling or mapping, a fixed-dimensional operation ticket feature vector is output. Example: In this embodiment, the fixed dimension is set to 256. That is, regardless of the length of the operation ticket text, it is ultimately mapped to a 256-dimensional feature vector. ∈R256, which facilitates unified retrieval later.

[0036] The step-by-step retrieval matching and similarity calculation are performed through the retrieval matching layer in the second deep learning model, based on the operation ticket feature vector. t is matched against the historical compliant ticket feature database. The historical compliant ticket feature database stores a massive number of historical feature vectors corresponding to historical compliant operation tickets. _1, 2,...}. The retrieval matching layer uses a cosine similarity and edit distance fusion algorithm to calculate the similarity score between the ticket to be reviewed and historical tickets. A specific calculation formula example is as follows: ,in, and Preset weighting coefficients (e.g.) =0.7, =0.3), EditDistance represents the text edit distance, and MaxLen is the maximum length of the two text segments. This fusion algorithm considers both the proximity in semantic space (cosine similarity) and the differences in text literal structure (edit distance).

[0037] The historical matching results are generated by selecting the top-K historical compliant tickets with the highest similarity scores as matching targets based on the calculated similarity scores. The pending operation ticket is compared with the matched historical compliant tickets to generate difference information. This difference information includes missing steps, differences in step order, and differences in device IDs. Finally, the historical matching results output by the second deep learning model are obtained. These historical matching results include: 1. Similarity score: Quantifies the degree of similarity between the invoice to be reviewed and historical compliance experience.

[0038] 2. Difference Information: Specific textual or logical differences described.

[0039] Through the above embodiments, this invention utilizes a composite architecture of a second deep learning model to achieve deep feature extraction of operation tickets. The BERT layer ensures the accuracy of semantic understanding, the BiLSTM layer solves the modeling problem of the temporal logic of operation steps, and the multi-head self-attention layer enhances the expressive power of key features. In addition, the fusion algorithm of cosine similarity and edit distance overcomes the defect that single semantic matching may ignore key literal differences (such as a difference of one digit in the device number), improves the reliability of historical matching results, and provides a solid historical empirical basis for the dual verification mechanism.

[0040] The second deep learning model is generated by training a database of historical compliant operation tickets as training data. It includes: obtaining a second training set of data, wherein the second training set of data includes multiple sets of second training data, and each set of second training data includes a first historical operation ticket, a second historical operation ticket, and labeled similarity and difference information between the first historical operation ticket and the second historical operation ticket. The second pre-set neural network model is trained based on the second training set data to generate the second deep learning model. The second pre-set neural network model is a composite neural network that uses BERT pre-training layers, BiLSTM sequence modeling layers, and multi-head self-attention layers.

[0041] Step S40: Generate an audit conclusion report based on the rule verification results and the historical matching results.

[0042] As an example, the similarity score is compared with a preset confidence threshold. For instance, a confidence threshold is preset in the decision fusion module. The preset confidence threshold measures whether the similarity between the operation ticket to be reviewed and historical compliance experience meets the confidence standard. In this embodiment, the preset confidence threshold is set to 0.85. The similarity score output by the second deep learning model is compared with the preset confidence threshold. Example: If the similarity score of an operation ticket to be reviewed is 0.92, it is determined to be greater than the threshold; if the score is 0.70, it is determined to be less than the threshold. When the rule verification result is a verification failure, the final review conclusion is determined to be a review failure, and the violation information is written into the review conclusion report. For example, when the rule verification result output by the first deep learning model is a verification failure, regardless of the similarity score of the historical matching results, the system determines that the final review conclusion is a review failure. At this time, the violation information in the rule verification result is directly written into the review conclusion report. The violation information includes the specific rule ID violated, the violation steps, and the reason for the violation. Example: If rule verification reveals a serious violation of disconnecting a switch under load, even if the operation ticket has a similarity of 0.99 with a historical ticket, it will still be deemed unqualified, and the report will indicate a violation of safety procedure R_002. This logic reflects the hard constraint principle in power audits that safety rules take precedence over historical experience. When the rule verification result is "verified and passed," and the similarity score is greater than or equal to a preset reliability threshold, the final audit conclusion is determined to be "approved." For example, when the rule verification result is "verified and passed," and the similarity score is greater than or equal to the preset reliability threshold (e.g., Score ≥ 0.85), the system determines the final audit conclusion to be "approved." At this time, the system writes the compliance information from the rule verification result into the audit conclusion report, and may attach the highly similar historical ticket number from the historical matching results as a reference. Example: If an operation ticket complies with all rules and has a similarity of 0.95 with a typical ticket in the historical compliance ticket database, the system determines it to be "approved," and the report indicates compliance with the procedure and high consistency with a typical historical ticket. When the rule verification result is "verification passed" and the similarity score is less than a preset reliability threshold, the final review conclusion is determined according to a preset conflict resolution strategy, which includes a rule priority strategy. For example, if the rule verification result is "verification passed" but the similarity score is less than a preset reliability threshold (e.g., Score < 0.85), it indicates that although the operation ticket does not violate existing rules, it differs significantly from historical routine operation patterns (possibly due to new equipment or new operating conditions leading to new compliance operations). In this case, the system determines the final review conclusion based on the preset conflict resolution strategy.

[0043] In this embodiment, the conflict resolution strategy adopts a rule-first approach. Specifically, since the rule verification has passed, it indicates that the operation ticket meets the mandatory requirements of the current procedure. Therefore, the final review conclusion is determined to be approved, but a low-risk warning or a suggestion for manual review will be marked in the review conclusion report. Example: The operation ticket steps of a newly commissioned substation are different from those of traditional stations, resulting in a historical similarity of only 0.60. However, the rule engine verification shows no violation of any procedure. The system determines that the operation ticket is approved based on the "rule-first approach," but the report indicates that the operation mode differs significantly from historical practices, and the special characteristics of the equipment should be confirmed. An review conclusion report containing the final review conclusion, compliance information or the aforementioned violation information, and difference information is generated. For example, by integrating the above processing results, a final review conclusion report containing the following content is generated: 1. Final review conclusion: Clearly indicate whether the review is passed or failed.

[0044] 2. Compliance or violation information: If approved, a list of compliant rules will be displayed; if not approved, details of the violation and suggested corrections will be displayed.

[0045] 3. Difference Information: Displays the differences between the pending audit ticket and the most similar historical compliant ticket (such as the addition or removal of steps, or the adjustment of the order).

[0046] Through the above embodiments, this invention achieves an organic combination of rigid rule constraints and flexible reference to historical experience. It ensures the absolute authority of safety procedures and prevents potential errors in historical data from misleading audits. The rule-priority strategy addresses the misjudgment problem caused by procedures updating faster than historical data accumulation, allowing new compliant operations to pass audits. Simultaneously, it prompts manual attention through discrepancy information, balancing the security and adaptability of the audit process. The final detailed audit conclusion report provides maintenance personnel with traceable decision-making basis, improving the intelligence level of operation ticket management.

[0047] In this embodiment, the structured review rules output by the first deep learning model are obtained by inputting the procedural text of the obtained operation ticket to be reviewed into the first deep learning model. The first deep learning model is trained using power regulations as training data and has the ability to extract rule logic from text. The operation ticket to be reviewed is then verified according to the structured review rules to obtain rule verification results. The obtained operation ticket text is then input into a second deep learning model to obtain historical matching results. The second deep learning model is trained using a historical compliant operation ticket database as training data and has the ability to obtain historical operation ticket similarity. Based on the rule verification results and the historical matching results, an review conclusion report is generated. This solves the technical problems of traditional review systems relying on static rules and having low data utilization in related technologies, achieving the effect of overcoming the limitations of static rule databases and improving data utilization.

[0048] Secondly, embodiments of this application also provide a device for reviewing operation ticket text.

[0049] In one embodiment, reference is made to Figure 2 , Figure 2 This is a schematic diagram of the functional modules of an embodiment of the device for reviewing operation ticket text in this application. Figure 2 As shown, the device for verifying the operation ticket text includes: The first acquisition module 10 is used to input the procedure text of the acquired operation ticket text to be reviewed into the first deep learning model to obtain the structured review rules output by the first deep learning model. The first deep learning model is generated by training using power procedure documents as training data and has the ability to extract rule logic from text. The second acquisition module 20 is used to verify the operation ticket to be reviewed according to the structured review rules and obtain the rule verification result; The third acquisition module 30 is used to input the acquired text of the operation ticket to be reviewed into the second deep learning model to obtain the historical matching results output by the second deep learning model. The second deep learning model is generated by training using a historical compliant operation ticket database as training data and has the ability to obtain the similarity of historical operation tickets. The generation module 40 is used to generate an audit conclusion report based on the rule verification results and the historical matching results.

[0050] Furthermore, in one embodiment, the first acquisition module 10 is used to: The procedure text of the obtained operation ticket to be reviewed is input into the first deep learning model; The encoder in the first deep learning model extracts professional entity information from the procedure text; The professional entity information is semantically parsed and information is extracted by the decoder in the first deep learning model to generate an intermediate logical representation. The intermediate logic representation format is converted by the rule compiler in the first deep learning model to obtain the structured audit rules output by the first deep learning model.

[0051] Furthermore, in one embodiment, the second acquisition module 20 is used for: The operation ticket to be reviewed is subjected to semantic structured parsing to extract entity information and generate key operation elements, wherein the entity information includes step sequence, device object and state verb; The key operational elements and the structured audit rules are judged by the pre-set rule verification engine to obtain the rule verification result, wherein the rule verification result includes pass and compliance information or fail and violation information.

[0052] Furthermore, in one embodiment, the third acquisition module 30 is used for: The text of the operation ticket to be reviewed is input into the second deep learning model; The semantic representation vector is obtained by semantically encoding the operation ticket to be reviewed through the BERT pre-trained layer in the second deep learning model; Temporal features are extracted from the semantic representation vector using the BiLSTM sequence modeling layer in the second deep learning model to obtain a sequence feature vector; By focusing key features on the sequence feature vector through the multi-head self-attention layer in the second deep learning model, a fixed-dimensional operation ticket feature vector is obtained. The retrieval and matching layer in the second deep learning model matches historical feature vectors based on the operation ticket feature vector, and calculates similarity scores using a cosine similarity and edit distance fusion algorithm to obtain historical matching results output by the second deep learning model. The historical matching results include similarity scores and difference information.

[0053] Furthermore, in one embodiment, the generation module 40 is used for: The similarity score is compared with a preset reliability threshold; When the rule verification result is a verification failure, the final review conclusion is determined to be a review failure, and the violation information is written into the review conclusion report; When the rule verification result is "verification passed" and the similarity score is greater than or equal to the preset reliability threshold, the final review conclusion is determined to be "review passed". When the rule verification result is in the verification passed state and the similarity score is less than the preset reliability threshold, the final review conclusion is determined according to the preset conflict resolution strategy, which includes the rule priority strategy. Generate an audit conclusion report that includes the final audit conclusion, the compliance information or the violation information, and the discrepancy information.

[0054] Furthermore, in one embodiment, the device for verifying the operation ticket text further includes a new module for: Obtain first training set data, wherein the first training set data includes multiple sets of first training data, and each set of first training data includes power regulations clauses and corresponding labeled structured rules; The first pre-set neural network model is trained based on the first training set data to generate a first deep learning model, wherein the first pre-set neural network model is DeBERTa-V3.

[0055] Furthermore, in one embodiment, the device for verifying the operation ticket text further includes a new module for: Obtain second training set data, wherein the second training set data includes multiple sets of second training data, and each set of second training data includes a first historical operation ticket, a second historical operation ticket, and labeled similarity and difference information of the first historical operation ticket and the second historical operation ticket; The second pre-set neural network model is trained based on the second training set data to generate a second deep learning model. The second pre-set neural network model is a composite neural network using a BERT pre-training layer, a BiLSTM sequence modeling layer, and a multi-head self-attention layer.

[0056] The functions of each module in the above-mentioned operation ticket text review device correspond to the steps in the above-mentioned operation ticket text review method embodiment, and their functions and implementation processes will not be described in detail here.

[0057] Thirdly, embodiments of this application provide a device for reviewing operation ticket texts. The device for reviewing operation ticket texts can be a personal computer (PC), a laptop computer, a server, or other devices with data processing capabilities.

[0058] Reference Figure 3 , Figure 3 This is a schematic diagram of the hardware structure of the operation ticket text review device involved in the embodiments of this application. In this embodiment, the operation ticket text review device may include a processor, a memory, a communication interface, and a communication bus.

[0059] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0060] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting internal components of the operation ticket text verification device, as well as interfaces used for interconnecting the operation ticket text verification device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0061] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0062] The processor can be a general-purpose processor, which can call the review program for the operation ticket text stored in the memory and execute the operation ticket text review method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the operation ticket text review program is called can refer to the various embodiments of the operation ticket text review method of this application, and will not be repeated here.

[0063] Those skilled in the art will understand that Figure 3 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0064] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0065] The present application provides a computer-readable storage medium storing a program for reviewing operation ticket text, wherein when the program for reviewing operation ticket text is executed by a processor, it implements the steps of the operation ticket text review method described above.

[0066] The method implemented when the operation ticket text review procedure is executed can be referred to in various embodiments of the operation ticket text review method of this application, and will not be repeated here.

[0067] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0068] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0069] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0070] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0071] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0072] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0073] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for reviewing operation ticket text, characterized in that, include: The procedure text of the operation ticket to be reviewed is input into the first deep learning model to obtain the structured review rules output by the first deep learning model. The first deep learning model is generated by training using power procedure documents as training data and has the ability to extract rule logic from text. The operation ticket to be reviewed is verified according to the structured review rules, and the rule verification result is obtained; The obtained text of the operation ticket to be reviewed is input into the second deep learning model to obtain the historical matching results output by the second deep learning model. The second deep learning model is generated by training using a historical compliant operation ticket database as training data and has the ability to obtain the similarity of historical operation tickets. Based on the rule verification results and the historical matching results, an audit conclusion report is generated.

2. The method for reviewing operation ticket text as described in claim 1, characterized in that, The step of inputting the procedure text of the obtained operation ticket to be reviewed into the first deep learning model to obtain the structured review rules output by the first deep learning model includes: The procedure text of the obtained operation ticket to be reviewed is input into the first deep learning model; The encoder in the first deep learning model extracts professional entity information from the procedure text; The professional entity information is semantically parsed and information is extracted by the decoder in the first deep learning model to generate an intermediate logical representation. The intermediate logic representation format is converted by the rule compiler in the first deep learning model to obtain the structured audit rules output by the first deep learning model.

3. The method for reviewing operation ticket text as described in claim 1, characterized in that, The step of verifying the operation ticket to be reviewed according to the structured review rules and obtaining the rule verification result includes: The operation ticket to be reviewed is subjected to semantic structured parsing to extract entity information and generate key operation elements, wherein the entity information includes step sequence, device object and state verb; The key operational elements and the structured audit rules are judged by the pre-set rule verification engine to obtain the rule verification result, wherein the rule verification result includes pass and compliance information or fail and violation information.

4. The method for reviewing operation ticket text as described in claim 1, characterized in that, The step of inputting the obtained text of the operation ticket to be reviewed into the second deep learning model to obtain the historical matching results output by the second deep learning model includes: The text of the operation ticket to be reviewed is input into the second deep learning model; The semantic representation vector is obtained by semantically encoding the operation ticket to be reviewed through the BERT pre-trained layer in the second deep learning model; Temporal features are extracted from the semantic representation vector using the BiLSTM sequence modeling layer in the second deep learning model to obtain a sequence feature vector; By focusing key features on the sequence feature vector through the multi-head self-attention layer in the second deep learning model, a fixed-dimensional operation ticket feature vector is obtained. The retrieval and matching layer in the second deep learning model matches historical feature vectors based on the operation ticket feature vector, and calculates similarity scores using a cosine similarity and edit distance fusion algorithm to obtain historical matching results output by the second deep learning model. The historical matching results include similarity scores and difference information.

5. The method for reviewing operation ticket text as described in claim 1, characterized in that, The rule verification result includes a verification pass status and corresponding compliance information, or a verification fail status and corresponding violation information; The historical matching results include similarity scores and difference information; the step of generating an audit conclusion report based on the rule verification results and the historical matching results includes: The similarity score is compared with a preset reliability threshold; When the rule verification result is a verification failure, the final review conclusion is determined to be a review failure, and the violation information is written into the review conclusion report; When the rule verification result is "verification passed" and the similarity score is greater than or equal to the preset reliability threshold, the final review conclusion is determined to be "review passed". When the rule verification result is in the verification passed state and the similarity score is less than the preset reliability threshold, the final review conclusion is determined according to the preset conflict resolution strategy, which includes the rule priority strategy. Generate an audit conclusion report that includes the final audit conclusion, the compliance information or the violation information, and the discrepancy information.

6. The method for reviewing operation ticket text as described in claim 1, characterized in that, The first deep learning model was generated using electricity regulation documents as training data, and includes: Obtain first training set data, wherein the first training set data includes multiple sets of first training data, and each set of first training data includes power regulations clauses and corresponding labeled structured rules; The first pre-set neural network model is trained based on the first training set data to generate a first deep learning model, wherein the first pre-set neural network model is DeBERTa-V3.

7. The method for reviewing operation ticket text as described in claim 1, characterized in that, The second deep learning model was generated using a historical compliant operation ticket database as training data, and includes: Obtain second training set data, wherein the second training set data includes multiple sets of second training data, and each set of second training data includes a first historical operation ticket, a second historical operation ticket, and labeled similarity and difference information of the first historical operation ticket and the second historical operation ticket; The second pre-set neural network model is trained based on the second training set data to generate a second deep learning model. The second pre-set neural network model is a composite neural network using a BERT pre-training layer, a BiLSTM sequence modeling layer, and a multi-head self-attention layer.

8. A device for verifying operation ticket text, characterized in that, The device for verifying the operation ticket text includes: The first acquisition module is used to input the procedure text of the acquired operation ticket text to be reviewed into the first deep learning model to obtain the structured review rules output by the first deep learning model. The first deep learning model is generated by training using power procedure documents as training data and has the ability to extract rule logic from text. The second acquisition module is used to verify the operation ticket to be reviewed according to the structured review rules and obtain the rule verification result; The third acquisition module is used to input the acquired text of the operation ticket to be reviewed into the second deep learning model to obtain the historical matching results output by the second deep learning model. The second deep learning model is generated by training using a historical compliant operation ticket database as training data and has the ability to obtain the similarity of historical operation tickets. The generation module is used to generate an audit conclusion report based on the rule verification results and the historical matching results.

9. A device for verifying operation ticket text, characterized in that, The operation ticket text review device includes a processor, a memory, and an operation ticket text review program stored in the memory and executable by the processor, wherein when the operation ticket text review program is executed by the processor, it implements the steps of the operation ticket text review method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for reviewing operation ticket text, wherein when the program for reviewing operation ticket text is executed by a processor, it implements the steps of the method for reviewing operation ticket text as described in any one of claims 1 to 7.

Images